v0.16.24

What's fixed

• Auth proxy hard timeout: supabase.auth.getUser() in the edge proxy is now bounded by a configurable timeout (default 2000ms, env-tunable). A stalled GoTrue or Supavisor pool can no longer cascade into MIDDLEWARE_INVOCATION_TIMEOUT 504s for every authenticated request; instead the request degrades to an unauthenticated state with an x-auth-degraded observability header.
• Webhook fail-fast: the Meta/WhatsApp webhook routes now return HTTP 503 with retry-after: 30 on transient DB errors (pool checkout timeouts, connection drops, fetch failures). Meta Cloud API retries any non-2xx with exponential backoff for ~7 days, so Meta's existing redelivery queue is now our durable buffer in front of Postgres — no more silent inbound-message drops during DB pool incidents.
• Structured incident logging: failed webhook ingestion logs a sanitized webhook.ingest.failed event (trace id, body sha256 hash, last completed pipeline stage, error name/message — no PII) plus a separate webhook.ingest.unexpected line for non-transient errors so Sentry alerts can distinguish logic regressions from infra incidents.

Why

Companion patch to v0.16.23. Where v0.16.23 narrowed the proxy matcher so unprotected API routes don't run Supabase auth at all, v0.16.24 adds a hard time bound around the auth call when it does run, and stops the inbound webhook pipeline from hanging when Postgres is sick.