Support large CD-ROMs in Windows 9x (MSF)

[blixt]

I had issues mounting a 300 MB ISO in Windows 98, and after a bit of digging it turned out it was because Windows wants the leadout in minutes:seconds:frames (MSF) format but we were providing it as blocks, so the larger the image the more wrong it was, causing misbehavior from the OS. (Sidenote: Windows NT/2000 will work with LBA and does not have this bug.)

For reference QEMU also does this conversion.

This should improve #1556, but I'm not sure if it fixes all the issues there.

As a "bonus" this PR also makes the device report "not present" when remounting an image since this seems to improve Windows 98 behavior.

[chschnell]

I tried this branch with Windows95b and it doesn't fix the problem described in #1556.

Could you please upload a short video demonstrating that you have a CD-ROM image > 255M working under Windows 98 (or is it 98 SE)?

We had already tried MSF encoding in #1556, and it didn't fix it for Win95 (I still have a branch atapi-toc-cmd). So it DOES fix it under Windows 98?

[blixt]

I tested many times with mounting the Office 97 ISO for Windows 98 (using your disk/state snapshot) and every time the OS didn't seem to recognize even the name of the disk and just became unresponsive when I tried to open D:, until I made the changes in this PR.

I have a some minor aesthetic custom HTML/CSS/JS on top of v86 running with the code in this PR in this recording:

vibe98-recording-github.mp4

The mount code I run is this:

async function mountCdrom(file: File) {
  const buffer = await file.arrayBuffer();
  emulator.eject_cdrom();
  await delay(1500);
  await emulator.set_cdrom({ buffer } as unknown as V86Image);
}

(It's an arrayBuffer() because I had some issues with lazy reads.)

[blixt]

And yes this is Windows 98 SE:

[copy]

Happy to merge if @chschnell gives this a green checkbox.

[chschnell]

Yes, this looks good to merge!

[copy]

(CI needs to be green though)