AudioStreamPlaybackWAV: Inherit from Resampled

[DeeJayLSP]

TL;DR: this is a non-breaking change that simplifies AudioStreamWAV a lot while improving the resampling quality.

A continuation and updated version of #83483, so everything that has been discussed there applies here. The original author, @wareya, has been added as the commit author, and I left myself as a co-author, as what I've done was mostly patching.

Removes the resampling code entirely from AudioStreamPlaybackWAV and makes it inherit from AudioStreamPlaybackResampled, fixes #58216. This brings the class in par with the Ogg Vorbis and MP3 equivalents, which already inherit from Resampled.

I have tested this with as many audio files I could, with every loop mode, and found no problems.

Any existing project needs no modifications to work with this PR. It barely modifies unexposed methods, so no compatibility breaks.

A few notable changes:

• IMA ADPCM audio now gets interpolated.
• do_resample did two things: convert samples from their format to AudioFrame, then apply linear interpolation. Interpolation code was removed and the function renamed to decode_samples.
• DATA_PAD has been removed from AudioStreamWAV as it lost its purpose. Functions that required handling the pads have been simplified, potentially reducing memory overhead.
• QOA caches were also removed, as they were a workaround workaround to make resampling work fine in the previous implementation. Unnecessary in this one.
• Linux release template builds using production=yes had a 4KB reduction in binary size. Note however that AudioStreamPlaybackResampled have an internal buffer of 1056B.

Caveats

Files imported with loop mode set to Forward, Ping-Pong or Backward in a build before #96768 was merged will cause crashes, as it will try to read one index out of bounds. It can be fixed by simply reimporting.

[DeeJayLSP]

Updated the offset variable back to int64_t. The max value can't be overflowed with a WAV file as those have a 32-bit size limit, and to overflow it with AudioEffectRecord it would take over 1.6 million years of recording (assuming 44100Hz stereo 16-bit).

Increment doesn't need to be multiplied by 8192, so I made it an int8_t as you only need 1 or -1.

[DeeJayLSP]

This technically fixes #58216, could this get a bug label?

[akien-mga]

Looks good to me from a cursory glance, and from reading the description. Let's get this tested.

[akien-mga]

Thanks!