Add SDL3 as a platform

[RobLoach]

This adds SDL3 as a platform to allow using the SDL3 File System when interacting with PhysFS. I'm not 100% sure this is a good idea, so let me know if this is a backwards approach.

Since the platform changes across SDL builds, I've exposed a PHYSFS_PLATFORM_SDL3 CMake option to allow explicitly stating that you want to target SDL3 instead of one of the other platforms.

• I confirm that I am the author of this code and release it to the PhysicsFS project under the Zlib license. This contribution does not contain code from other sources, including code generated by a Large Language Model ("AI").

[icculus]

I've actually been thinking of doing this for awhile, honestly.

I'll look at this more closely in the morning (and probably merge it).

[RobLoach]

Sounds good with me. I ran it through my own test suite, and things seem to be working, but feel free to change anything around. Some thoughts...

1. Enumeration system could likely improve with further safety checks, but seemed okay to me
2. I wasn't 100% on if SDL_GetPrefPath() issues a SDL_SetError(), so that's the only SDL call that doesn't do an SDL_BAIL_IF()

Overall though, it seemed to be pretty solid.

[icculus]

SDL_GetUserFolder, SDL_GetBasePath, and SDL_GetPrefPath all promise the returned string will end with a path separator, so this function probably can go away in favor of a call to __PHYSFS_strdup().

[RobLoach]

The only difference I found between them was that on SDL_GetPrefPath(), we don't need to call __PHYSFS_strdup() since it allocs the memory for us, while the others don't. Should be good now. Thanks!

[icculus]

I think it's fine to pollute SDL_GetError(). If one didn't query the string immediately upon return from a failing SDL API call, it's not reasonable to assume it didn't get replaced by something else (including a PhysicsFS that was explicitly built to use SDL under the hood).

[RobLoach]

Removed SDL_BAIL_IF entirely. Does clean it up.