OpenAL audio implementation

[OmniBlade]

Implements an OpenAL audio back end.

WIP, has issues with object life times.

[caseychaos1212]

From Codex:

Findings

1. The OpenAL cache is not equivalent to the Miles cache, and it can hold dangling pointers. m_BufferCache stores raw FFMpegBufferClass* keyed by const char* pointer identity, and Get_Sound_Buffer() inserts the buffer without taking a cache-owned ref. The caller then drops its own ref immediately, so once the sound object/handle release theirs, the delayed-release thread can delete a buffer that the cache still points at. That matches the author’s symptom. Relevant code: OpenALAudio.h, OpenALAudio.cpp, WWAudio.cpp, soundhandle.cpp.

2. Even if the cache owned a ref, FFMpegBufferClass is stateful and therefore unsafe to cache/share between sounds. The cached object contains the decoder state (FFmpegFile m_FileHandle) and a mutable staging buffer, and every handle mutates that shared object via Reset_Buffer() / Refresh_Buffer() while queueing audio. Two simultaneous plays of the same filename will fight over decode position and PCM contents. Relevant code: FFMpegBuffer.h, FFMpegBuffer.cpp, OpenALHandle.cpp, OpenALHandle.cpp.

The delayed-release/cache issue the author mentioned is real, but the main difference from Miles is structural: Miles caches immutable buffer data and its cache entry owns a ref; this OpenAL version caches a mutable decoder object and the cache does not own it. I did not run a build or in-game test on the branch; this was a static review of PR #103.

[Dadud]

I am an AI agent. Static code review of PR #103.

Bugs:

1. FFMpegPlayer.cpp Render() — null check removed, potential crash

   // Was removed:
   if (CurrentFrame->data == nullptr) {
       return;
   }

   GotFrame is set on the first decoded frame of any type, but CurrentFrame->data can be null at construction before the first video frame is rendered. This causes a null dereference in Render(). Suggest restoring the check.

2. OpenALHandle.cpp destructor — SampleUsers map entry never removed

   OpenALHandleClass::~OpenALHandleClass()
   {
       alSourcei(SampleHandle, AL_BUFFER, AL_NONE);
       alDeleteBuffers(OAL_BUFFER_COUNT, SampleBuffers);
       // Missing: SampleUsers.erase(SampleHandle);
   }

   SampleUsers is a static unordered_map. Entries are written by Set_Sample_User_Data() but never erased. When a handle is destroyed, the map retains a stale entry. If OpenAL recycles the ALuint name, subsequent lookups resolve to the wrong live object.

Reliability issues:

3. OpenALAudio.cpp — Flush_Cache() is empty

   Shutdown() calls Flush_Cache() before closing the ALC device. If any FFMpegBufferClass instances hold resources not explicitly freed in the handle destructor, they will leak.

4. OpenALHandle.cpp — Initialize() doesn't reset SampleBufferIndex

   On handle reuse, SampleBufferIndex retains its prior value. Queuing then starts from the wrong buffer slot on subsequent plays.

Design note (not a blocker):

Get_Sound_Buffer() creates a new FFMpegBufferClass per call. This avoids the mutable-state-sharing issue Miles has, but differs from Miles's caching behavior. Documented for completeness; no action required.

[OmniBlade]

I am an AI agent. Static code review of PR #103.

Bugs:

1. **FFMpegPlayer.cpp Render() — null check removed, potential crash**     
   GotFrame is set on the first decoded frame of any type, but CurrentFrame->data can be null at construction before the first video frame is rendered. This causes a null dereference in Render(). Suggest restoring the check.

CurrentFrame->data is an array, not a pointer, in the version of ffmpeg I am building against so a nullptr check against it is tautological false and the compiler will ellide it anyhow.

2. **OpenALHandle.cpp destructor — SampleUsers map entry never removed**      
   SampleUsers is a static unordered_map. Entries are written by Set_Sample_User_Data() but never erased. When a handle is destroyed, the map retains a stale entry. If OpenAL recycles the ALuint name, subsequent lookups resolve to the wrong live object.

This is handled by AudibleSoundClass which sets the associated object to NULL when freeing its handle.