index: persist and retry webhook deliveries

[shamsartem]

Summary

• persist each webhook request in a dedicated Room queue before network delivery
• retry transport failures and HTTP 408, 425, 429, and 5xx responses; resume pending deliveries when the app restarts
• expose manual Retry for permanent HTTP failures without re-running transcription or the agent
• send a stable X-Index-Delivery value across attempts so receivers can deduplicate ambiguous retries
• clear audio, transcription, URL, and headers from successful queue rows while retaining the delivery-ID tombstone

Fixes #336. The background queue and stable event identifier also address the retry/idempotency requests discussed in #305.

Delivery semantics

Delivery is at-least-once. A timeout after the receiver accepted a request can cause another attempt, so receivers should treat X-Index-Delivery as an idempotency key and return success for a duplicate. The normal recording operation completes after the payload is durably queued; webhook retries do not repeat transcription or agent work.

This is built on the current per-gesture webhook model. #302 touches several of the same webhook files. #373 also introduces X-Index-Delivery as part of HMAC signing; if it lands first, this branch will need a small rebase so the stable queue ID is fed into its signing path.

Tests

• :experimental:testAndroidHostTest
• :experimental:compileAndroidDeviceTest
• :androidApp:assembleDebug (using the repository dummy Firebase config)

The queue and transport code live in commonMain for Android and iOS. I could not build the Apple target on Linux.

AI disclosure

This change was developed with AI assistance. I reviewed the complete diff, the queue state transitions, retry classification, database migration, payload cleanup, UI path, and documentation, and ran the tests and Android builds listed above.

[CLAassistant]

All committers have signed the CLA.