Fix crash in FileAccessCompressed and improve error handling

[RedworkDE]

Fixes partially #62585 (on MSVC #74582 is also required or it will just crash in CowData::resize before any of this matters)

Only the change in WRITE_FIT is strictly required to fix the crash.
The error handling improvements are required so that ResourceSaver::save doesn't OK when output file will be corrupted, with them it correctly returns ERR_CANT_CREATE.
Without the FileAccessCompressed::store_buffer implementation, the file writing will instead get stuck spamming errors for each byte in the buffer.

[Calinou]

The behavior of this PR rebased against master a83eb16 is the same as in the linked issue on Linux: no crash, but I still see the error message:

ERROR: Condition "p_size < 0" is true. Returning: ERR_INVALID_PARAMETER
   at: resize (./core/templates/cowdata.h:262)

test.res weighs 544,482,622 bytes when I exit the MRP after the error message is printed.

[RedworkDE]

Other than the resize crash on MSVC, the crash caused by this might be ASAN only iirc.

Also the main difference is that with the PR ResourceSaver::save returns ERR_CANT_CREATE instead of OK when it fails. The save failing can't be fixed without redesigning how the compressed files work or adding support for 2GB+ arrays (and even then this PR is still useful as saving can still fail in some cases where it then still shouldn't return OK)

[TokisanGames]

With PR #86730 merged, Godot still crashes in #62585. Applying this PR on top, it replaces the crash with an error popup. However large resource files are still not saved. See #62585 (comment) for details.

[akien-mga]

The changes look good to me.

I'll unlink #62585 as it won't be fully fixed yet, but this seems to be a step in the right direction.

[akien-mga]

Would be worth rebasing as this was last updated a year ago.