Display reasoning effort in the web dashboard

[jrob5756]

Problem

The dashboard already streams reasoning content (chain-of-thought shows up as 💭 thinking in the activity stream), but it does not show the configured reasoning effort (low / medium / high / xhigh) for each agent. Users have had to consult the YAML or logs to confirm which tier was applied.

What this changes

Adds a Reasoning row to each agent's metadata grid in the detail panel when reasoning effort is set in YAML. It appears right after Model so they're visually grouped.

Effective value per agent (no source distinction in the UI):

1. agent.reasoning.effort if set, otherwise
2. workflow.runtime.default_reasoning_effort if set, otherwise
3. omitted (MetadataGrid filters null/empty rows automatically)

Implementation

Backend

• engine/workflow.py — the workflow_started agent payload now includes reasoning_effort. Subworkflows inherit the same emission path via their own engine instance, so a single change covers both top-level and nested workflows.
• New tests in test_event_emission.py cover per-agent override, workflow-default fallback, and unset (None).

Frontend

• types/events.ts — WorkflowStartedData.agents shape gains reasoning_effort.
• stores/workflow-store.ts — NodeData, IterationSnapshot, and WorkflowAgent types each gain reasoning_effort. The workflow_started handlers (top-level + subworkflow) copy the field onto the node, mirroring the existing model assignment. The agent_started snapshot push also carries it through to per-iteration history.
• components/detail/MetadataGrid.tsx — buildAgentMetadata renders a Reasoning row right after Model when present.
• components/detail/AgentDetail.tsx — the current-iteration snapshot spread passes reasoning_effort through so the live and history views match.
• Static bundle rebuilt (npm run build).

Notes

• Replays of older event log JSONL files just won't render the row — the field is missing, so MetadataGrid skips it gracefully. No migration needed.
• Reasoning effort is fixed per workflow run (not per turn), so reusing the same value across iteration history snapshots is correct.

Verification

• ruff check / ruff format --check — clean
• Full test suite: 2374 passed / 9 skipped, no regressions
• Frontend tsc -b && vite build — clean

Closes the second of the two reasoning-display issues uncovered while wiring through the recent reasoning.effort work.