Add support for observability + external optimizer

[aliroberts]

Summary

Adds weco observe CLI commands for tracking external optimization runs. The CLI is write-only — it creates runs and logs steps. Lifecycle (close/archive) is managed by the dashboard.

New commands

• weco observe init — Create a run and print the run ID to stdout for capture via $().
  • -s/--source or --sources — source files to track (matches weco run convention)
  • --metric — primary metric name
  • -g/--goal — maximize/max/minimize/min (matches weco run convention)
  • --name, -i/--additional-instructions — optional metadata
• weco observe log — Log a step with metrics, code snapshot, and optional tree lineage.
  • --run-id, --step (required)
  • --status completed|failed, --description, --metrics (JSON), --parent-step
  • -s/--source or --sources — re-reads files from disk

Design

• Fire-and-forget — all commands print warnings to stderr on failure but always exit 0. Never crashes the agent's loop.
• No lifecycle commands — complete/fail were removed. External runs are perpetually active; the dashboard handles closing. Logging a step to a closed run silently reopens it.
• Naming conventions — flags match weco run (--source/--sources, --goal, --additional-instructions).

Files

File	Purpose
weco/observe/cli.py	Argument parsing + command handlers
weco/observe/api.py	HTTP client (create_run, log_step)
weco/observe/observer.py	Python SDK (WecoObserver, ObserveRun)
weco/observe/__init__.py	Public exports
weco/cli.py	Registers observe subcommand