binaryfail

They're actually mascot characters for an online store called DLsite, not from a game or anything as far as I know.

Engine dev here. Extra1 isn't using opus at all, so the mem leak shouldn't be a problem this time around. (It was used on Aokana's Kickstarter physical release and Switch port though, so a small patch wouldn't hurt if needed.)

Does CircusEditor have a way to add a line break or did you have to paste it in another way?
For the 91xx remapping, I used a hex editor to add a 0x91 byte in front of each accented letter. This was easy to do manually since it's just one or two sentences for my example text.
I started with typing it as regular text, viewed as codepage-1252
   Olá, este é um teste Add a 91 byte in for each accented letter, viewed as sjis
   Ol黛, este 鷹 um teste Convert the remaining letters to full-width in sjis
   Ｏｌ黛，　ｅｓｔｅ　鷹　ｕｍ　ｔｅｓｔｅ ...and then paste the resulting text into the script. It should be possible to automate the procedure above for your translation. One way would be to add support for remapping accented letters within CircusEditor -- it is open-source.
My dll also remaps the full-width alphabet back to ASCII so the letters look normal instead of  ｗｉｄｅ  in the game, and it was usually necessary when non-Japanese fonts don't have support for sjis codepoints.

Accented characters: The square symbols is probably because the character was misinterpreted as something else by the font. What tool are you using to edit the game's MES files? This will ultimately need to be customized to work appropriately with whatever way you're handling accented letters in your script + font.
When I did accented letters for my example, I essentially remapped accented symbols to the 91xx SJIS range and the dll would map them back to the intended codepoint. For example, ê is 0xEA in unicode/cp1252 and outputted as 91EA in the script file.
Line breaks: the game expects the script to provide explicit line breaks in the text using the 0A byte. That means your editing tool needs to be able to insert line breaks into the text; the game will not auto-break lines by itself.

Most of the font overrides I did were handled with a dll that I wrote. Here's my current one for it, just put it in the same folder as the game exe.
Finally, the text spacing is a hex edit on the exe. Since I did my testing on the disc release of EXS, these next instructions will only work for that version.
Use a hex editor and look at address 1CA32 in the exe, and you should see the bytes A1 B8 52 BB 00 ..... now overwrite those with B8 0C 00 00 00 where 0C is the amount of spacing. Something between 09 and 0C should suffice, depending on the font.
I should mention that a bunch of technical considerations will need to be kept in mind, but I'll let you try it first and see how that goes.

The font name can be found in the exe (in this case, it is ＭＳ 明朝) which can be patched via a hex edit, or overridden with other means. But that's the easy part.
For Portuguese, you also need to have some way to use accented letters, which is something not natively supported in SJIS encoding. There are some ways to handle that, such as an edited font, or patching in support to the exe.
This game's engine doesn't accept normal ASCII text in the script, so you're forced to use full-width letters which  ｍａｋｅｓ　ｅｖｅｒｙｔｈｉｎｇ　ｌｏｏｋ　ｗｉｄｅ in the game. That also can be fixed by patching the exe, which seems to have a hardcoded amount for the text's letter spacing.
A decent looking starting point could look like this. The new font being used here is Comic Sans MS, of course.

The font name can be found in the exe (in this case, it is ＭＳ 明朝) which can be patched via a hex edit, or overridden with other means. But that's the easy part.
For Portuguese, you also need to have some way to use accented letters, which is something not natively supported in SJIS encoding. There are some ways to handle that, such as an edited font, or patching in support to the exe.
This game's engine doesn't accept normal ASCII text in the script, so you're forced to use full-width letters which  ｍａｋｅｓ　ｅｖｅｒｙｔｈｉｎｇ　ｌｏｏｋ　ｗｉｄｅ in the game. That also can be fixed by patching the exe, which seems to have a hardcoded amount for the text's letter spacing.
A decent looking starting point could look like this. The new font being used here is Comic Sans MS, of course.

Yes. At the very least, you should use PONScripter-fork as the engine. For the script: decrypt it with nsdec, re-save the text as UTF-8, and replace the ` symbol with ^ for text lines because of PONScripter syntax.

Now I guess a few things will need to be changed, based on what I'm seeing.
You're translating to French, so Shift-JIS text encoding won't work. We want the script to be stored as unicode (like UTF-8), but the existing ONScripter build doesn't seem to read UTF-8. Use a different engine such as PONScripter-fork: https://github.com/sekaiproject/ponscripter-fork/releases
Get the file named narci2-windows.exe (find it under the "Assets" label) and put it into the game's folder. You will use this exe to play the game.
I made a few changes to the script to run with PONScripter. Get the file here and put 0.utf in the game folder
At this point, the text should look like this.
You can translate the text in 0.utf and play without needing to encrypt it again, so just forget about nsdec and nscript.dat from now on.
Next is the font, the file called default.ttf which can be replaced with something else. Here's what it looks like with Arial.
(PS: the "pointers" you mentioned are called accents. Shift-JIS can't encode them, so we needed unicode.)

Ah, you installed the game into Program Files, so doing anything in there would've required admin access.
You should copy the game folder to somewhere you have regular write permissions, such as your desktop. From there, I expect the tool to work normally.

This is what I got for the result.txt file: link here.
And for what it's worth, I did try it on Windows 8.1 which worked fine. (I ran the second command twice to test one more edit.)

I tried making a workaround for this issue. Can you test this module and see if that fixes it?

The start of zlib data should be easy to look for. Just find the bytes 78 9C in this case.
In the .mwb file, the value at 0x14 is the uncompressed size and the value at 0x18 is the compressed size. The data itself starts at 0x1C.

The game seems to use the Inconsolata font by default. Perhaps not having that installed might explain things.

Try this fix: download link
Seems like the engine has a numerical bug where having more than 2GB of VMEM gets read as a negative amount, thus having <64MB. I guess some other Lune games will also have this bug.

Try this module here.
It adjusts the font width down by 20% so the text won't overlap as badly as it normally would.

I've seen this issue before with another Cyberworks game. Use this...
 

I've seen this issue before with another Cyberworks game. Use this...
 

Try this module here.
It adjusts the font width down by 20% so the text won't overlap as badly as it normally would.

Try this module here.
It adjusts the font width down by 20% so the text won't overlap as badly as it normally would.

After the initial extraction from the ws2, the tool gives you a text file and a bin file. In most cases, you can leave the bin file alone. Here's a how-to if you want it.

I tried making a workaround for this issue. Can you test this module and see if that fixes it?

Hex edit the exe at offset 0xB84B7 changing 03 to 01. That should allow the game to run without needing to set your calendar format.
It's arg1 to MultiByteToWideChar, btw. As a bonus idea: if you're able to make it push 0x3A4, that should allow the game to run regardless of your locale+date setting.

Hex edit the exe at offset 0xB84B7 changing 03 to 01. That should allow the game to run without needing to set your calendar format.
It's arg1 to MultiByteToWideChar, btw. As a bonus idea: if you're able to make it push 0x3A4, that should allow the game to run regardless of your locale+date setting.

I tried making a workaround for this issue. Can you test this module and see if that fixes it?