feat(analytics): record how long a successful AI generation takes

[Guotai812]

Summary

• New PostHog event ai_task_generation_succeeded, fired once per AI task generation turn that returns a result (ReceiveGenerationResult):
  • duration_ms: from mic release (voice) or the send tap (text) to the final result, timed on the phone.
  • transcript_ms (voice only): same start to ReceiveTranscript, to tell slow transcription from slow generation.
  • task_count: every draft shown after the turn (tasks + recurring tasks + notes), including drafts kept from earlier turns. Same rule as onboarding_voice_task_generated.task_count.
• The voice clock now starts at mic release in useVoiceRecorder.stopAndUpload; before, it started after the audio file was read and base64-encoded. ai_task_generation_failed.duration_ms uses the same clock.
• Code-review fixes in the same functions:
  • Voice send failures (AudioSubmitFailed) now carry duration_ms.
  • A voice network drop after the transcript arrived is labelled stage generation, not transcription.
  • A connection drop after a result or error already closed the request no longer records a second, failed outcome or shows an error toast.
• Analytics impact: voice failure durations now include the recorder stop and upload prep, and some voice NetworkError failures move from transcription to generation. Compare old and new numbers by app version.
• Mobile only, no backend change. After merge: check on TestFlight and Android alpha that voice and text still produce drafts (the recording hand-off changed).

Verified: local run on the iOS simulator against the local API. PostHog is off outside production, so the payloads were printed with a temporary log line (not committed):

voice                        ai_task_generation_succeeded {"input_mode":"voice","duration_ms":5776,"task_count":1,"transcript_ms":574}
text                         ai_task_generation_succeeded {"input_mode":"text","duration_ms":7857,"task_count":1}
voice, API stopped mid-run   ai_task_generation_failed {"input_mode":"voice","stage":"generation","error_code":"NetworkError","duration_ms":1975}
voice, API down              ai_task_generation_failed {"input_mode":"voice","stage":"send","error_code":"AudioSubmitFailed","duration_ms":19}

5 voice turns and 1 text turn each sent exactly one success event; the failed turns sent none. Phone totals line up with the server's model time (text turn: 7857 ms on the phone, 7783 ms for the model call). tsc 0 errors, expo lint no new warnings.

Release note

none

Status:

• User-facing — announce it
• Beta / partial — announce, but tagged as beta
• Hidden in production (feature-flagged / not enabled for users) — don't announce
• Internal only (refactor / infra / tests / CI / deps) — don't announce

[github-actions]

Test Results

145 tests   145 ✅  11s ⏱️
  1 suites    0 💤
  1 files      0 ❌

Results for commit 01729ac.

[github-actions]

Package	Line Rate	Branch Rate	Complexity	Health
BlotzTask	4%	36%	2957	❌
BlotzTask	4%	36%	2957	❌
Summary	4% (5086 / 117230)	36% (1124 / 3166)	5914	❌