feat(recorder): on-device recorder host + whisper transcription + chat/RAG

[dishit-wednesday]

TL;DR - Open-core host that wires in the private pro/ recorder submodule and ships
the shared on-device transcription + retrieval engine it relies on. Recorder feature code
stays closed-source in pro/; this PR is the public wiring + shared services. Pairs with
off-grid-ai/mobile-pro#11 (the submodule pointer references a commit on that pro branch, so
they land together).

flowchart TB
  subgraph pro["pro/ submodule (private, autolinked)"]
    rec["Recorder + AudioNormalizer"]
  end
  subgraph core["core (public host)"]
    slot["MemoryTabScreen<br/>paywall + registry slot"]
    ws["whisperService<br/>windowed transcribe + CoreML guard"]
    rag["RAG 'Recordings' KB<br/>metadata column + migration"]
    chat["chat: summarize attachment"]
  end
  slot -. "registered-screen slot" .-> rec
  rec -->|"16k mono WAV"| ws
  ws -->|"transcript + t0/t1 segments"| rag
  ws -->|"transcript"| chat

Loading

What's in this PR

• Recorder host - Memory tab -> Recorder; Settings moved to Home header; guarded pro
  native autolink (react-native.config.js); iOS background-audio mode.
• Whisper - windowed / resumable / memory-bounded transcription; CoreML asset guard;
  .en language forcing; iOS streaming-lifetime patch (whisper.rn+0.5.5).
• Chat / RAG - recordings knowledge base (chunk metadata column + migration);
  summarize oversized transcript/document attachments.
• Integration - .gitmodules re-pointed to canonical off-grid-ai/mobile-pro;
  submodule pointer bumped to the merged pro commit.
• Temporary - reset-license testing controls (remove/re-gate before shipping).

How it works (expand for detail)

Transcription pipeline (whisper.cpp via whisper.rn)

• Windowed, resumable, memory-bounded. transcribeFile drives whisper.rn with an
  {offset, duration} window so long recordings transcribe in chunks and resume from a
  checkpoint (last finished chunk) instead of reloading the whole file. Live cumulative
  segments surface through whisper.rn's onNewSegments (native new_segment_callback),
  wired only when nProcessors == 1 (parallel decode cannot stream in order).
• iOS streaming lifetime patch (whisper.rn+0.5.5). The file path
  (transcribeFile -> ObjC transcribeData) crashed because new_segment_callback's
  user_data was a stack struct freed before the callback fired. The patch hoists it so it
  outlives the call; compiled into the iOS binary, so live segments work on both platforms.
• CoreML asset guard. Enabling CoreML without the per-model
  ggml-<model>-encoder.mlmodelc bundle makes whisper.rn fail CoreML init, fall back to
  CPU, then crash at transcribe 0% on some devices (A12 / iPhone XS). loadModel checks the
  encoder asset exists and downgrades useCoreMLIos to false when missing, so the toggle is
  a no-op instead of a crash.
• English-only language forcing. .en models carry only English tokens; auto-detect
  (returns codes like "tg") or an explicit non-English pick makes whisper unstable on iOS
  (crash at 0%, 762s for 13%, garbled). transcribeFile forces language: 'en' for any
  model whose catalogue lang === 'en', falling back to the .en.bin filename convention.

Retrieval (recordings knowledge base)

Transcripts index into a local RAG "Recordings" KB. rag_chunks gains a metadata TEXT
column (recordingId, startMs, eventTitle, ...) via a guarded ALTER TABLE migration for
older installs; chunk metadata is JSON-serialized on write.

Integration / build

• The pro native library is autolinked through a guarded react-native.config.js (only
  when the pro android/iOS sources are present), so closed-source code never leaks into the
  public repo but builds when the submodule is checked out.
• .gitmodules re-pointed from the old wednesday-solutions/offgrid-pro to the canonical
  off-grid-ai/mobile-pro; submodule pointer bumped to the merged pro commit.

Testing

npm run lint (0 errors), tsc --noEmit (clean), and the full jest suite pass; the
pre-push hook ran them, no --no-verify used. Added CoreML-fallback + RAG-metadata coverage.

Follow-ups (not blocking)

• Slot-key indirection: MemoryTabScreen looks up the recorder by the concrete pro
  screen name; move it behind an abstract slot key so public never names a pro internal.
• whisperService max-lines: deferred with a TODO; clearing it means moving model
  download/management into its own module (~11 call sites across core + pro).
• Walkthrough video on the PR.

Reviewer's guide (commits by theme)

The branch is chronological; this groups the commits by area.

• Recorder host + navigation - rename Memory tab to Recorder; Memory tab (Pro
  recorder/paywall) + move Settings to Home header; autolink pro native + pod install; iOS
  background-audio mode.
• Whisper transcription - core file-transcription; offset/duration chunking + live
  segments + CoreML; CoreML asset guard + .en forcing + tdrz; iOS streaming lifetime
  patch; memory-snapshot logging; whisper remote-server provider type.
• Chat / RAG - recordings KB (chunk metadata, indexText, seed project); summarize
  oversized attachments; stream summaries live.
• Integration (submodule) - 4x bump pro submodule; point submodule at canonical
  off-grid-ai/mobile-pro.

[sonarqubecloud]

Quality Gate passed

Issues
22 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
1.0% Duplication on New Code

See analysis details on SonarQube Cloud