feat(galaxy): add LLMCallEvent and CostThresholdExceededEvent to event bus

[nuthalapativarun]

Feature: LLM Cost & Token Usage Tracking

UFO already computes cost per LLM call in ufo/llm/base.py and returns it from every provider, but the value is discarded by callers. This PR series wires that existing data through the event bus, aggregates it in the metrics observer, exposes it via a FastAPI router, and visualises it in the Galaxy Web UI.

This is PR 1 of 2. PR 2 (full backend wiring + frontend) depends on this PR and will be linked once this is reviewed.

---

What this PR does

Adds the event type definitions that the rest of the feature depends on — no runtime behaviour changes.

galaxy/core/events.py

• Adds LLM_CALL_COMPLETED and COST_THRESHOLD_EXCEEDED to the EventType enum
• Adds LLMCallEvent(Event) dataclass with fields:
  • agent_type: str — which agent made the call (e.g. HOST_AGENT)
  • model: str — model name (e.g. gpt-4o, claude-3-5-sonnet-20241022)
  • prompt_tokens: int
  • completion_tokens: int
  • cost: float — estimated USD cost
  • duration_ms: float — wall-clock time of the API call
• Adds CostThresholdExceededEvent(Event) dataclass with fields:
  • session_id: str
  • total_cost: float
  • threshold: float

ufo/trajectory/parser.py

• Replaces bare except: with except json.JSONDecodeError: to avoid masking unexpected exceptions during trajectory evaluation file parsing.

---

Architecture

LLM Provider (openai / claude / gemini)
  └─ returns (response, cost, tokens)
       └─ llm_call.py: emits LLMCallEvent          ← wired in PR 2
            └─ EventBus
                 └─ SessionMetricsObserver          ← wired in PR 2
                      └─ MetricsService
                           └─ /api/metrics/*        ← added in PR 2
                                └─ CostDashboard    ← added in PR 2

---

Testing

Pure data structure addition — no logic changes, no new dependencies. Existing tests unaffected.