feat: github rest client for dataset proposals

[Scr4tch587]

What changed

New `core/github/` package: a minimal GitHub REST client (`GitHubClient`) used by the upcoming dataset-proposal endpoint to open PRs with a bot token. Methods:

• `get_branch_sha` / `commit_files_to_new_branch`: commits a set of files as a single commit on a new branch via the git data API (tree → commit → ref)
• `create_pull` + `request_reviewers`: opens the PR and requests reviewers best-effort — the PR author is filtered out (GitHub refuses author-as-reviewer), and if a batch request fails (e.g. one login isn't a collaborator) each reviewer is retried individually so one bad entry can't drop the rest; failures log warnings and never fail the flow
• `open_pr_with_files`: the one-call orchestration the service layer uses
• errors surface as `GitHubError` (carries HTTP status); an existing proposal branch raises the `BranchAlreadyExistsError` subclass (422 "already exists")
• config from env: `GITHUB_TOKEN` (required), `GITHUB_REPO`, `GITHUB_API_URL` (override used to e2e-test against a local fake GitHub)

Built on `requests` (already a production dependency). Tests mock at the transport-adapter layer — no network.

Why

Dataset creation is being added as a PR-based flow: submissions become pull requests so code review stays the gate for anything that executes on the server (per the trust model in SPEC-backend). This client is the transport for that. First PR of the dataset-creation stack (on top of the react migration stack).

Benefit

One clean commit per proposal, reviewer auto-assignment that degrades gracefully, and a client that's fully testable offline.

🤖 Generated with Claude Code

[github-actions]

• main
  • feat: replace svelte frontend with react + tailwind scaffold #375
    • feat: typed api client and api key handling #376
      • feat: dataset browsing ui #377
        • ci: deploy frontend to github pages #378
          • docs: update specs for react migration #379
            • feat: github rest client for dataset proposals 👈