fix(agent): capture_pprof reports worker-wedge gauges + 30s capture throttle (#2422)

[ToddHebebrand]

Summary

Fixes both gaps in the capture_pprof diagnostic command (#2394/#2408 seam):

1. Wedge gauges were always 0/0. handleCapturePprof discarded its *Heartbeat receiver and embedded collectors.CollectRuntimeStats() directly, which never populates commandsInFlight/commandsOverdue. The snapshot now goes through h.collectAgentRuntime(now) — the wrapper that wires in inFlightCommandStats — so a capture taken while chasing an overdue-commands heartbeat trend reflects the real pool state.

2. No throttle on GC-forcing captures. Heap/all captures unconditionally ran runtime.GC(); as a server-queued command (10 concurrent / 100 queued), a burst could force back-to-back stop-the-world GCs. Added a 30s minimum interval between admitted captures using the same atomic CAS slot pattern as the heartbeat watchdog dump throttle (fix(agent): stop heartbeat watchdog from firing every heartbeat on slow links (#2386) #2392). Throttled captures fail with an explicit rate-limited error. Payload validation runs before the throttle so malformed requests don't consume the slot.

All changes are confined to handlers_diag.go / handlers_diag_test.go — heartbeat.go is untouched.

Tests

• New: TestHandleCapturePprofIncludesWedgeGauges (tracked overdue command shows up as 1/1 in the snapshot), TestHandleCapturePprofThrottled (second capture rejected, slot frees after interval), TestHandleCapturePprofValidationDoesNotConsumeSlot.
• Existing diag tests updated to use a zero-value *Heartbeat + per-test throttle reset.
• cd agent && go test -race ./internal/heartbeat/... — ok (41s); go build ./... clean; go vet clean.

Closes #2422

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying breeze with    Cloudflare Pages

Latest commit:	c429baf
Status:	✅  Deploy successful!
Preview URL:	https://100990ce.breeze-9te.pages.dev
Branch Preview URL:	https://fix-2422-capture-pprof-gauge.breeze-9te.pages.dev

View logs

[ToddHebebrand]

Review run: /pr-review-toolkit:review-pr (code-reviewer, pr-test-analyzer, silent-failure-hunter)

Findings: 0 blocking across all three agents. 4 non-blocking items raised → 3 addressed in c429baf:

• Concurrency test for capturePprofTryAcquire (the CAS loop is a copy of the watchdog's, so the watchdog's concurrency test didn't cover it) — added TestCapturePprofTryAcquireConcurrent (32 racers, exactly one winner).
• Interval boundary pin — added TestCapturePprofTryAcquireInterval (1ns-before rejected, exact interval admitted).
• Rate-limit error wording claimed every capture forces a GC; goroutine-only captures are throttled but never call runtime.GC() — message now says "heap captures".

Declined (with rationale):

• Partial results when the heap profile exceeds the size cap (return gauges + goroutine profile alongside a heapProfileError): a behavior change to pre-existing all-or-nothing semantics, beyond [Agent] capture_pprof reports zero worker-wedge gauges and lacks a capture throttle #2422's scope. The failure is loud, and profile=goroutine works after the interval.
• Wall-clock skew guard in the throttle (now-last < 0 after a backward NTP step keeps rejecting until wall time catches up): identical characteristic exists in the fix(agent): stop heartbeat watchdog from firing every heartbeat on slow links (#2386) #2392 watchdog throttle this mirrors; fixing one copy would diverge the pattern. Worth a small follow-up covering both if it matters in practice.

Tests: cd agent && go test -race ./internal/heartbeat/... — ok (11 CapturePprof tests, full package green); go vet + go build ./... + gofmt clean. CI Test Agent passed on the first commit; re-running on c429baf.

Status: review-clean, awaiting maintainer merge.