perf(slack): dedupe webhook JSON.parse + LOG_LEVEL=debug perf timings

[JonasJesus42]

Summary

Two small perf items spun out of the latency investigation in #442:

A. Eliminate the duplicate JSON.parse

verifySlackRequest was re-deserializing rawBody after the router had already parsed it. Two JSON.parse per webhook on the same payload. The function now returns a plain boolean and the router reuses its existing parsedPayload for shouldIgnoreEvent + processConnectionEventAsync. Saves ~1 ms and removes a dead code branch.

B. Perf-timing log gated on LOG_LEVEL=debug

receivedAt = Date.now() is now captured at the top of the webhook POST handler in router.ts and threaded through:

router → processConnectionEventAsync
       → SlackEventContext.{ traceId, receivedAt }
       → handleSlackEvent
       → per-handler HandlerMeta arg

Each chat-style handler (handleAppMention / handleDirectMessage / handleThreadReply, plus the triggerOnly fast path) emits two markers via logger.debug:

• perf:thinking_sent — ms from webhook receipt to chat.postMessage of "Pensando..." resolving.
• perf:trigger_published — ms from webhook receipt to trigger published.

logger.debug is dropped below the configured min level (see logger.ts:50 — if (LOG_LEVELS[level] < LOG_LEVELS[this.minLevel]) return), so these only flow to HyperDX when LOG_LEVEL=debug is set on the pod. Production runs (LOG_LEVEL=info, the default) pay only the Date.now() + branch cost.

Combined with the existing [Slack] sendMessage SUCCESS in Xms + the thread_messages count on the publisher's notify log, an operator can now decompose end-to-end webhook latency into pod overhead, Slack postMessage RTT, and trigger-enrichment time from logs alone — without code changes.

Test plan

• Type-check + build clean (already green locally).
• Deploy with LOG_LEVEL=info (default) → no perf:* entries in HyperDX. Bot still answers as before. Webhook latency should be marginally better (~1 ms from parse dedup).
• Flip the pod to LOG_LEVEL=debug → perf:thinking_sent and perf:trigger_published markers appear in HyperDX with duration_ms and trace_id. Correlate against [Slack] sendMessage SUCCESS in Xms for pod-vs-Slack breakdown.
• Send a DM, an app_mention, and a thread reply → all three handlers emit both markers.
• triggerOnly mode emits perf:trigger_published from handleSlackEvent directly (no thinking_sent marker since no Pensando is sent in that mode).

---

Summary by cubic

Removes duplicate JSON parsing in the Slack webhook path and adds debug-gated perf timing logs for chat handlers. This trims ~1 ms per webhook and lets operators break down latency from logs without code changes.

• New Features

  • Debug perf markers via logger.debug: perf:thinking_sent and perf:trigger_published for app mentions, DMs, thread replies, and the triggerOnly path; emitted only when LOG_LEVEL=debug (no impact at LOG_LEVEL=info).

• Refactors

  • Deduplicated webhook parsing: verifySlackRequest now returns a boolean; the router reuses its parsedPayload for shouldIgnoreEvent and processConnectionEventAsync.
  • Captured receivedAt in the router and threaded through SlackEventContext (with traceId) to support timing correlation.

^{Written for commit 2ee57b8. Summary will update on new commits.}