2133: Add microphone spoofing media stack + libmicspoofing

[RankoR]

Implements GrapheneOS/os-issue-tracker#2133

• AudioRecord fallback to spoofed source on PERMISSION_DENIED when mic spoofing is enabled for UID
• AAudio input disables MMAP for spoofed UIDs
• MediaRecorderClient allows audio source when spoofing is enabled
• introduces Rust/C FFI libmicspoofing with default/custom WAV source loading, conversion, and safety fallbacks
• adds instrumentation, native, Rust tests, and fuzzer

Tests:

atest -c \
  MicSpoofingTests \
  MicSpoofingGrantedTests \
  libmicspoofing_rs_test \
  mic_spoofing_tests \
  aaudio_spoofed_test \
  opensles_spoofed_test

Fuzzing:

m mic_spoofing_fuzzer
adb sync data
FUZZ_BIN="$(adb shell 'find /data/fuzz -type f -name mic_spoofing_fuzzer 2>/dev/null | head -n 1' | tr -d '\r')"
adb shell "$FUZZ_BIN -max_total_time=60"

[RankoR]

1. There are buffer re-allocations each call in some branches. Due to the nature of the hardened allocator, it's "costlier" than on the default allocator, where its cost is near-zero. However, it's still not too "expensive", and spoofing shouldn't happen too often to affect user-perceived performance. On the other hand, if we pre-allocate a buffer and reuse it, it would make code & memory management more complex, introducing new risks. IMO, this small tradeoff is ok.
2. As we're using a silent WAV by default, it makes testing more difficult (as tests expect non-zero samples). So there are hacks to replace the default silent file with a non-silent one in tests.