feat(sdk): TypeScript SDK — crash-resume via resumeRun - #103

Adds the resume path to the TypeScript SDK so a Node/TS agent can pick its work
back up after a crash — the JS counterpart to the Python SDK's resume_run.

Runtime.resumeRun(runId, fn) re-attaches to an existing governed run by id. Unlike
governedRun, it neither creates a new run nor cancels on error: the daemon reloads
non-terminal runs on restart with the budget and usage they had already spent, so
enforcement continues without re-spending. The agent reads run.latestCheckpoint()
to resume from where it left off — the run id is the only thing it has to keep across
a restart.

await rt.resumeRun(runId, async (run) => {
  const cp = await run.latestCheckpoint();
  const start = (cp?.payload?.cursor as number) ?? 0;
  for (let i = start; i < total; i++) {
    await run.step();                     // counts against the SAME budget
    await run.checkpoint("step", { cursor: i + 1 });
  }
});

Tests

Exercised over real HTTP against the in-process mock daemon (no daemon, no keys):

• a run that crashed after 3 paid steps re-attaches, reads its checkpoint, and
  continues stepping at index 3 (the counter does not reset) — no overspend on restart;
• resumeRun never cancels a run it didn't start when the body throws;
• an unknown run id surfaces APIError(404).

Docs: a "Resume after a crash" section in the SDK README and a CHANGELOG entry.

Closes #80.