Update docs for 0.1.6: board-less submit, mrg jobs, decoupled job status

[cscaff]

Summary

• docs/guides/cli.md and docs/guides/sdk.md still described the pre-0.1.6
  client-side board picking (find_idle_fpga, --fpga-id/App(fpga_id=...)
  as a way to request a board for a fresh build) and the old
  job/logs/cancel <fpga_id> [job_id] CLI signature. Rewrote all of it to
  match the real board-less build-slot pipeline: fpga_id is discovered from
  the completed job, not picked; --fpga-id only matters for the
  --no-program reconnect case; job/logs/cancel take a bare job_id;
  added the missing mrg jobs command.
• docs/guides/quickstart.md: fixed the "picks an idle board automatically"
  line to match.
• Added a ## 0.1.6 changelog entry.
• docs/reference/cloud.md: fixed the mkdocstrings member list --
  NoFPGAAvailableError no longer exists in mrg.cloud (removed along with
  find_idle_fpga), and Stream was missing despite being a real, current
  export. mkdocs build --strict passes.

Separate finding, not fixed here

src/manhattan_reasoning_gym/bench/silicon.py (CloudSilicon, the
sandbox-to-real-silicon broker backend) still calls _client.find_idle_fpga()
and the old _client.submit(fpga_id, path, ...) signature -- both removed in
the board-less refactor. This is dead code today (raises AttributeError the
moment it runs) with zero test coverage, and isn't part of reference/bench.md's
documented member list, so it's a real SDK bug rather than a docs issue. Flagging
it here rather than fixing it silently -- happy to open a follow-up PR for it if
wanted.

Test plan

• mkdocs build --strict (with _code/ refreshed to the real 0.1.6
  source) -- no warnings or errors
• Manually cross-checked every rewritten CLI/SDK claim against the actual
  _cli.py/_app.py/_client.py source (argparse definitions, docstrings)

🤖 Generated with Claude Code