feat: add dual-track recording to capture each side of a call separately

[R2bEEaton]

Summary

Adds a "Record each side separately" toggle (Settings -> Audio configuration) that captures the uplink (your mic) and downlink (the other party) as two independent, time-synced files instead of one mixed recording, e.g. call_..._uplink.opus + call_..._downlink.opus.

• ShellService now runs a second, independent ShellAudioPipeline instance behind two new AIDL methods (startSecondaryRecording/stopSecondaryRecording), so a second scrcpy-server process (voice-call-downlink) runs concurrently alongside the existing one (voice-call-uplink), each with its own socket/pipe.
• AudioRecordingEngine was reworked around a private TrackSession bundle so it can drive one (normal) or two (dual-track) capture pipelines in parallel, with symmetric start/rollback/teardown for both.
• ScrcpyAudioMuxer accepts an optional shared wall-clock origin so both files' PTS=0 line up to the exact same instant, keeping the two tracks in sync for downstream processing.
• RecordingFileNameFormatter gained an optional filename suffix (_uplink/_downlink) to distinguish the two outputs.
• When the toggle is on, the Audio source dropdown is hidden (the uplink/downlink sources are forced) instead of shown disabled.

Why this matters

Today, recording both sides of a call only produces one mixed file (VOICE_CALL), or a single-sided file (VOICE_CALL_UPLINK/VOICE_CALL_DOWNLINK) if you pick one manually — there's no way to get both sides at once without them being mixed together. For transcription workflows, a mixed track makes speaker attribution unreliable (diarization heuristics), whereas two separate, perfectly-synced mono files make "who said what" deterministic: it's just "which file". This mirrors how call-center transcription products (e.g. Twilio/Deepgram dual-channel) typically separate speakers.

Two files (rather than one file with two audio tracks) was chosen deliberately: Android's MediaMuxer only supports a single audio track per OGG container, so a true single-file two-track approach would only work for AAC/MP4, not Opus. Two mono files work identically for both codecs, need no new audio pipeline stage, and can still be losslessly combined into one stereo file afterward with a single ffmpeg command (documented in docs/configuration.md) if a specific transcription API requires that shape instead.

Testing

• Verified end-to-end on a real device (Pixel 10 Pro, Android 16) via Shizuku: placed and received real calls with the toggle off (unchanged single-file behavior, confirmed byte-identical code path) and on (confirmed two files appear in the recording folder with _uplink/_downlink suffixes, both playable, both non-empty, both starting from the same wall-clock instant).
• Confirmed pause/resume and mid-call cancel/error paths correctly tear down both tracks with no orphaned scrcpy-server process and no partial/locked output file.
• Confirmed toggling the setting on/off correctly hides/reveals the Audio source dropdown.
• Full ./gradlew assembleDebug build passes with no new warnings from the changed files.

This is additive and off by default; existing single-track recording behavior is unchanged when the toggle is off.

[R2bEEaton]

Apologies — I should have opened an issue first to discuss this before submitting a PR, per the contributing guidelines (large features should be discussed before the work is done). Closing this for now; I'll open an issue instead so we can discuss the approach first.