v0.5.0

Highlights

• Trace engine - record every receive()/send() as structured traces
  with nanosecond timestamps. Enable with a single flag:

  inspector = Inspector(app, trace=True)
  # ... drive the app ...
  record = inspector.traces[0]
  record.scope.method      # "GET"
  record.scope.path        # "/api/users"
  record.summary.ttfb_ns   # time to first byte (ns)

  Pluggable storage via TraceStorage protocol. Built-in: MemoryStorage
  (default) and FileStorage (one JSON file per trace).

• Deterministic sampling - sample_rate parameter on Inspector.
  Hash-based, same endpoint always produces the same decision for a given rate.

• asgion trace CLI command - human-readable trace output with
  color-coded phases, inline violation markers with severity, and per-record
  breakdown:

  $ asgion trace myapp:app
  

  TRACE  GET / (0.070ms, TTFB 0.042ms)
  
       0.020ms  send     http.response.body  4 bytes  ← HF-002 (error)
       0.042ms  send     http.response.start  200  (+0.022ms)  ← SEM-002 (info)
       0.059ms  send     http.response.body  5 bytes  (+0.016ms)
  
    Events: 3  |  Violations: 2 (1 error, 1 info)
  

  --format json for machine-readable output, --out for file storage.

Fixes

• CLI check/trace: WS runner now sends a websocket.receive message
  before disconnect, avoiding app crashes on frameworks that expect data
  after accept (e.g. Starlette).

• CLI check: summary line now reports application errors instead of
  showing "No violations found" when a scope raised an exception.

---

Full changelog: CHANGELOG.md