addpkg(main/oboe) + enhance(main/openal-soft, main/pulseaudio): Oboe audio stack

[ferrumclaudepilgrim]

Summary

Adds Oboe 1.10.0 as a new package and wires
it into openal-soft and pulseaudio.

Oboe is Google's C++ library for high-performance audio on Android, wrapping
AAudio (API 26+) and OpenSL ES with automatic fallback.

Commit 1: addpkg(main/oboe)

• Builds shared library only (BUILD_SHARED_LIBS=ON)
• Includes hand-written pkg-config and CMake config files (upstream does not
  provide install(EXPORT) or .pc templates)

Tested on 2 devices:

• Samsung S26 Ultra (Android 16): load, unload, find_package, pkg-config,
  linking stress, all PASS
• Google Pixel 10 Pro (Android 17 Beta): same suite, all PASS

Commit 2: enhance(main/openal-soft)

Three changes to packages/openal-soft/build.sh:

1. TERMUX_PKG_REVISION=1
2. oboe added to TERMUX_PKG_DEPENDS
3. -DALSOFT_REQUIRE_OBOE=ON in TERMUX_PKG_EXTRA_CONFIGURE_ARGS

Upstream's existing Oboe backend (alc/backends/oboe.cpp) compiles in via
find_package(oboe CONFIG). No patches needed.

A/B comparison on both devices:

• Oboe backend: initialized, played audio, 48kHz stereo, 0 XRuns
• OpenSL ES backend: fails with undefined symbol: swift_demangle (the same
  dlopen issue from fix(main/openal-soft): disable dlopen() of opensles for luanti #23149 and ALSOFT_DLOPEN=ON ALSOFT_REQUIRE_OPENSL=ON builds not working at runtime inside Termux on random Android devices kcat/openal-soft#1111)

Commit 3: enhance(main/pulseaudio)

Adds module-oboe-sink and module-oboe-source. TERMUX_PKG_REVISION
bumped from 1 to 2.

Tested on Samsung S26 Ultra (Android 16) and Google Pixel 10 Pro (Android 17
Beta). Multi-arch build clean across aarch64, arm, i686, x86_64.

Relates to #28861, #27978, #27367, #26871.

[robertkirkman]

This works for me and it fixes the problem with openal-soft on certain devices.

[ferrumclaudepilgrim]

Landed commit 3 per @robertkirkman's invite on #29074. PR title and body
updated to cover all three commits.

Evidence for commit 3:

• Samsung S26 Ultra (Android 16 / API 36): sink + source PASS (audio by ear,
  real mic capture, 3-cycle suspend/resume, 5-cycle load/unload)
• Google Pixel 10 Pro (Android 17 Beta 4 / API 37): same suite, all PASS
• Multi-arch build in termux-package-builder Docker: aarch64 / arm / i686 /
  x86_64, all exit 0

SLES remains the active default; Oboe modules are commented-out options in
default.pa.

[shiumano]

When Termux is running in the background, plugging in or unplugging headphones causes pulseaudio to freeze.
(Incidentally, aaudio also has the same problem of freezing when headphones are connected or disconnected.)
You'll likely encounter situations where this happens frequently, such as when music is playing in the background or when Termux:X11 is open.

[ferrumclaudepilgrim]

Sorry for the delay, I've been moving and it's been... fun.

@shiumano thanks for surfacing this. I dug into it pretty heavily. I was able to reproduce the issue in question using BT headphones as that is all I have currently. While definitely a thing it predates this PR based on my research. It is in module-aaudio-sink.c, the existing AAudio sink. The error_callback function calls pa_sink_suspend directly from AAudio's error thread, which is not a thread PulseAudio owns.

I reproduced it on my S26 Ultra by playing a 5 minute sine wave, backgrounding Termux, then disconnecting my headphones. PA went into deadlock almost immediately. pactl info still answered meaning main thread was good.

The new Oboe modules in this PR inherit the same pattern, character for character. So the bug shape is shared across multiple modules.

I see a fix: replace the direct suspend with pa_asyncmsgq_post to the sink (or source) IO thread, handle the suspend on that thread, drop the spin-wait. My brain says that's a different cup of tea for a different day since the already widely scoped PR doesn't get any more ambitious.