Retry failed Index webhook deliveries

[Hades32sBot]

Summary

• Webhook uploads were fire-and-forget: a non-2xx response or network error was logged to run history and permanently dropped. This adds in-memory retries on transient failures — network errors/timeouts, HTTP 408, 429, and 5xx — with 1s/2s/5s backoff (3 retries).
• Other 4xx responses and signing errors still fail immediately (retrying can't help them), and sendTestEvent stays single-shot so the settings UI gets instant feedback.
• Adds a retryable flag to IndexWebhookRunResult and an isRetryableStatus predicate, covered by a new unit test.

Note: this is deliberately in-memory only — it covers brief connectivity blips but not a dead endpoint or killed app process. A persisted retry queue (in the style of MemfaultChunkQueue) could be a follow-up if wanted.

Test plan

• New IndexWebhookRetryTest covers the retryable-status predicate and backoff bounds
• Manual: configure webhook to an endpoint returning 503, then 200 — verify retry succeeds and run history records the final result
• Manual: configure webhook to unroutable host — verify ~8s of retries then FAILED in run history

cc @Hades32

[CLAassistant]

All committers have signed the CLA.