fix(audio): retain completed AudioOut2 grains until pushed

[eljackjoestar-cmd]

What

During a level-loading test, Lost Soul Aside (PPSA26358) queued more than sixteen completed audio grains before Push consumed them. sceAudioOut2ContextAdvance then threw AudioOut2: more than 16 completed grains await a push, terminating the process.

Remove the emulator's sixteen-grain restriction. Completed PCM and gain snapshots remain queued in order until sceAudioOut2ContextPush consumes them. The regression advances 64 distinct grains on a producer thread using a rewritten buffer, ends that thread, then pushes and checks every stereo output sample. The technical-debt note records that an unconsumed backlog retains PCM storage and that a console backlog limit is unverified.

Tested

• Windows 11, WinLibs GCC 15.2, Release. audio_out2_port_buffers fails with the new regression and main's implementation, throwing at the seventeenth advance; it passes with this change (1.78 s). Rebased onto current main 0afa6a607ed0; all 10 focused audio checks pass again (7.27 s).
• All 10 audio_out2_* and audio_out_shutdown checks pass (3.80 s), covering PCM order, rewritten buffers, gain snapshots, cross-thread pushes, output timing and shutdown.
• Title A/B: the same local integration on main a077a88df, with existing perf(agc): import resident direct memory through a section alias on Windows #5851, fix(agc): promote a held import window to its whole allocation #5870 and perf(agc): allow shader preparation on first use #5872 (lazy preparation enabled), a missing-metadata PlayGo diagnostic and -onethread, using the same relinked executable. Before this change the run aborted in AudioOut2 during loading; with this change it passes that point. The user confirmed gameplay in run173; a later run with the same verified executable/module hashes (run177) also reached gameplay, captured through the game window with the combat tutorial visible. The cutscene ran around 6 FPS and the later scene around 1 FPS, with substantial memory pressure; sustained playability is not established. Run173 was stopped by the old stdout logging guard, not a demonstrated allocation failure. This does not resolve the separate garbage-collection timeout seen with the normal threaded launch. perf(agc): import resident direct memory through a section alias on Windows #5851 has since been closed in favor of perf(agc): keep Windows host imports of write-watched private memory watched #6060; the recorded integration above retained the tested perf(agc): import resident direct memory through a section alias on Windows #5851 version. Runtime diagnostics and other integration changes are absent from this PR.
• tools/check_conventions.py --base origin/main and git diff --check pass.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw; no silent stub added
• No investigation files or images added to the repository
• New third-party code is a submodule built from source: n/a
• Generic behaviour, not specific to one title
• Depends on: none
• Unblocks: PPSA26358 Lost Soul Aside (PPSA26358): bring-up and remaining blockers #4401 (the AudioOut2 backlog abort; gameplay entry observed in the disclosed integration, performance remains poor)
• GPUs tested: n/a (audio change)
• AI-assisted: yes

[Dextroid17]

I ran the test case this PR adds, on Linux.

Setup: head 6575aa2 merged into main at ff23657 (no conflicts; the PR's base is 262 commits behind), Linux, default CMake build (CMAKE_BUILD_TYPE empty, unoptimized), clean environment (env -i with only HOME, PATH and VK_ICD_FILENAMES set). ctest -R '^audio_out2_port_buffers$', 3 runs each.

• with this PR: 3 of 3 runs pass.
• with core/libs/prx/libSceAudioOut/src/AudioOut2Context.cpp put back to main's version (the PR's test changes kept): 3 of 3 runs fail, each aborting on main's 16-grain limit (uncaught std::runtime_error: "AudioOut2: more than 16 completed grains await a push").

The test uses SDL's disk audio driver, so no sound device was involved. I did not run Lost Soul Aside.
I did not check Windows or macOS.

AI-assisted: yes

[isnisk]

Checked on Linux: Ubuntu 26.04 (WSL2), GCC 15.2.0, Release, this head (6575aa217) merged onto main 489466d3f (clean merge).

• With the change: audio_out2_port_buffers and the other 16 audio_out* tests pass, 3 rounds, including the four audio_out*_shutdown tests from fix(libSceAudioOut): park guest audio calls at exit before the port tables are destroyed #6000; no compiler warnings.
• Without it (main's AudioOut2Context.cpp, the new test kept): audio_out2_port_buffers aborts 3 of 3 with what(): AudioOut2: more than 16 completed grains await a push.
• check_conventions.py --base origin/main passes.

Not checked: macOS, the title.

AI-assisted: yes (Claude Code).

[github-actions]

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Pull request	Shared files not compared
PR 5021	docs/dev/debt/libSceAudioOut.md
PR 6447	docs/dev/debt/libSceAudioOut.md
PR 6493	docs/dev/debt/libSceAudioOut.md
PR 6964	docs/dev/debt/libSceAudioOut.md