Add hackathons CLI commands + MCP/CLI parity gate

[kaggle-agent]

Ships the four hackathon MCP tools as kaggle hackathons (alias h)
verbs for the 2.2.0 release: get, writeups list/download/resolve-links.
Also adds a CI gate (tools/check_mcp_cli_parity.py) that fails when a
new MCP tool ships without a corresponding CLI command or an explicit
skip reason, so future drift is caught at PR time rather than at release.

---

Task: mooneyp-20260504232100-ef35439c
Context: https://chat.kaggle.net/kaggle/pl/poh4xrmnrffi8dzfmz9qzxdapw

[paultimothymooney]

See relevant feature request with additional context here: https://buganizer.corp.google.com/issues/509657800

[kaggle-agent]

Re: @paultimothymooney's review:

Thanks for the pointer. I don't have access to the internal buganizer link, but I'll treat the proposal in the PR description as the source of truth for scope. If there's additional context from b/509657800 that should land in the plan (e.g., specific hackathon UX requirements, additional MCP tools beyond the four listed, or a different priority on the parity gate), please paste the relevant bits here and I'll fold them into .kagent/plans/hackathons-cli-and-mcp-parity.md before implementation begins.

[paultimothymooney]

@sloppy please implement the plan from https://github.com/Kaggle/kaggle-cli/pull/996/changes

[kaggle-agent]

Re: @paultimothymooney's review:

This comment is directed at @sloppy to implement the plan in a follow-up PR — this PR (#996) only contains the design doc itself. No code changes are needed here; the implementation work (hackathons CLI commands + parity gate) will land in a separate PR by @sloppy based on .kagent/plans/hackathons-cli-and-mcp-parity.md. Leaving this PR as plan-only.

[paultimothymooney]

The purpose of this PR should not be to add a planning document to the repo. The purpose of this PR is supposed to be to actually implement the plan that is currently described at https://github.com/Kaggle/kaggle-cli/pull/996/changes. Please investigate and resolve.

[kaggle-agent]

Re: @paultimothymooney's review:

Implemented the plan in this PR. Changes:

Hackathons CLI (alias h) — 4 commands wired into src/kaggle/cli.py and src/kaggle/api/kaggle_api_extended.py:

• kaggle h get <competition> → get_hackathon_overview
• kaggle h writeups list <competition> → list_hackathon_write_ups
• kaggle h writeups download <competition> [-p path] → paginates and writes CSV (default ./<competition>-writeups.csv)
• kaggle h writeups resolve-links <competition> <writeup_id> → surfaces WriteUp.write_up_links via get_hackathon_write_up

Tests: src/kaggle/test/test_hackathons_cli.py (20 tests, all passing).

MCP↔CLI parity gate: tools/check_mcp_cli_parity.py (stdlib-only ast walker that follows parse_* helpers across function boundaries), tools/mcp_cli_mapping.yaml seeded with 67 tools, tools/README.md, and .github/workflows/parity.yml.

Docs: skills/SKILL.md updated, new skills/references/hackathons.md, CHANGELOG entries under ### Next.

Verification: 20/20 hackathons tests pass; python tools/check_mcp_cli_parity.py exits 0; ruff introduces 0 new errors. One pre-existing test failure (test_benchmarks_cli.py::test_writes_all_vars) left alone per scope rules — it asserts a stale LLMS_AVAILABLE list unrelated to this PR.

Plan deviations (driven by what the published kagglesdk 0.1.23 actually exposes):

• resolve-links requires both <competition> and <writeup_id> positionals because ApiGetHackathonWriteUpRequest is scoped by competition; no standalone WriteupsClient.get_resolved_writeup_links RPC exists yet.
• download is implemented client-side because no download_hackathon_write_ups RPC exists in the SDK.
• The parity script's --mcp-client-url defaults still point at the proposed raw URL, but it falls back to mapping-only validation (with a notice) when the fetch fails — currently the case because kaggleazure isn't publicly fetchable. CI sets GITHUB_TOKEN so the fetch will succeed once the file is reachable.