fix(dashboard,pipeline): hydrate cost/latency + barge-in gate from first audio

[nicolotognoni]

Summary

Two regression-class bugs surfaced during 0.6.0 acceptance against releases/0.6.0/typescript/matrix/outbound-cartesia-cerebras-elevenlabs.ts:

• Dashboard hydrate wiped cost / latency / caller / callee for every call rebuilt from disk → all hydrated rows showed \$0.00 and —.
• Pipeline early barge-in self-cancelled the agent's first turn before any TTS chunk reached the wire → 0 bytes emitted, line silent.

Fixes are minimal, parity-aligned (Py + TS), and fully backward-compatible.

Implementation

1. MetricsStore.hydrate schema mismatch

CallLogger.log_call_end writes cost/latency/duration_ms/telephony_provider as top-level keys of metadata.json. The reader looked under meta.metrics.cost / meta.metrics.latency. New helper metricsFromTopLevel / _metrics_from_top_level synthesizes a metrics dict from those top-level fields when meta.metrics is missing, mapping latency.p95_ms → metrics.latency_avg.total_ms so the existing UI fields populate. Explicit meta.metrics (legacy/future shape) is preserved untouched.

2. Pipeline barge-in gate anchored on actual audio emission

Cloud TTS first-byte latency is 200-700 ms; the 250 ms PSTN gate measured from _begin_speaking / beginSpeaking expired BEFORE TTS produced audio. New _first_audio_sent_at / firstAudioSentAt is set in _mark_first_audio_sent / markFirstAudioSent invoked AFTER bridge.sendAudio / audio_sender.send_audio succeeds at the four pipeline emit sites (firstMessage, streaming response, regular response, WebSocket remote). _can_barge_in / canBargeIn now returns false while the marker is null. Gate values (250 ms / 1000 ms) unchanged — only the anchor moves.

Files:

• libraries/python/getpatter/dashboard/store.py
• libraries/python/getpatter/stream_handler.py
• libraries/typescript/src/dashboard/store.ts
• libraries/typescript/src/stream-handler.ts

Breaking change?

No — defaults preserved, all existing fields/methods unchanged. The barge-in fix is a behavioural improvement (was already documented as a "anti-flicker only" gate; the anchor change makes that contract actually hold).

Test plan

• Python: pytest tests/ → 1717 passed, 7 skipped
• TypeScript: npm test → 1394 passed, npm run lint clean, build clean
• New regressions:
  • test_hydrate_lifts_top_level_cost_and_latency_into_metrics (Py)
  • test_hydrate_preserves_explicit_metrics_when_present (Py)
  • test_barge_in_suppressed_before_first_audio_emitted (Py)
  • MetricsStore.hydrate > lifts top-level cost/latency/duration into metrics (TS)
  • MetricsStore.hydrate > preserves explicit metrics when present (TS)
  • barge-in gate > canBargeIn() false before the first TTS chunk has hit the wire (TS)
• Existing _handle_barge_in / handleBargeIn tests updated to set both timestamps for the new contract.

Docs updates

N/A — bug fix in internal behaviour. CHANGELOG ## Unreleased updated with both entries.

[nicolotognoni]

Superseded by #85 (0.6.1 release bundle). All commits from this branch (fix/dashboard-hydrate-schema-and-bargein-grace) have been merged into feat/observability-otel-attrs-0.6.1, plus the new opt-in barge-in confirmation strategies (MinWordsStrategy, BargeInStrategy Protocol/interface). Closing without merge — please review #85 instead.