Parallel execution-tree fetch, request hooks and client-side rate limiting (v0.1.34)

[Volv-G]

get_execution_details used to fetch a run's execution tree with one sequential request per execution. It now fetches level by level in parallel, without changing its output, and the client paces its own requests.

• Parallel tree fetch: new tangle_cli.execution_tree.fetch_execution_tree fetches breadth-first, one batch per level, with up to execution_tree_max_workers (8) threads. It runs inline without threads (e.g. Pyodide). get_execution_details uses it, and the enriched output is unchanged.
• Request hooks: no-op prefetch(paths) / persist(paths) on TangleApiClient, as a generic caching extension point.
• Client rate limiting: max_requests_per_second (default 8, None disables, off under emscripten) paces every physical send across threads.
  • A 429 or 503 pauses the shared limiter; the other workers resume with jitter.
  • A 503's Retry-After is now honoured, including for log-stream opens.
  • Limiter waits never carry a send past the retry deadline.
  • Auth refresh and header snapshots are serialized across workers.
• Version: 0.1.34. The README documents the fetcher, hooks and limiter.

Tests

• Run-details goldens generated from the previous sequential traversal (byte-identical output; each execution fetched once).
• Fetcher, hook and threading tests.
• Deterministic fake-clock rate-limiter tests, plus client-integration tests.
• Full suite passes locally.