docs: document dataset proposal flow in specs

[Scr4tch587]

What changed

• SPEC-backend.md: new "Dataset proposals endpoint" section — `POST /datasets` request shape, the six validation steps (byte-identical config re-validation, registry cross-checks, AST + ruff lint of the builder script), PR shape (branch naming, single commit, reviewer auto-request), env configuration (`GITHUB_TOKEN` / `GITHUB_REPO` / `GITHUB_REVIEWERS` / `GITHUB_API_URL`), full status-code table, and the trust-model rationale (code review stays the gate; the manual-`.env` gap is documented). `POST /datasets` added to the public endpoint list.
• SPEC-frontend.md: the "Planned" placeholder becomes the real "Create dataset" feature section — form sections (proposer identity, schema/dependency editors, CodeMirror), two-layer validation, lazy loading; navigation, file structure, dependency, and API-integration sections updated (`proposeDataset`, error `detail` extraction, MutationCache 401 handling).

Why

AGENTS.md requires spec files to track code changes; this documents the dataset-creation stack (#383–#386) and closes out the stack.

🤖 Generated with Claude Code

[github-actions]

• docs: document dataset proposal flow in specs 👈
  • feat: api client support for triggering builds #388