docs: llms.txt, provider-request template, contribution ladder

[jothimani-rajendran]

What

Adds the contributor-plumbing pieces this repo was missing: llms.txt at the repo root, a
provider_request.md issue template for requesting a new context-report run/judge model
backend, and the shared "ladder" section plus the seeded good-first-issue search link in
CONTRIBUTING.md. No code, README, or workflow changes.

• llms.txt: modeled on agentseam's and chock's, listing the docs an assistant should read to
  understand this project (README, CONTRIBUTING, CONVENTIONS, AGENTS, SECURITY, GOVERNANCE,
  CHANGELOG, the two spec READMEs, docs/cli.md, docs/library.md, the measurement paper) —
  every path verified to exist.
• .github/ISSUE_TEMPLATE/provider_request.md: this repo's analogue of an "adapter" is a model
  provider behind PROVIDERS in src/context_report/run/runner.py (the Asker protocol a
  backend implements — see openai_backend.py/cli_backend.py), not a per-agent hook adapter,
  so the template asks for exactly that: how a prompt is sent and answered, auth, token-usage
  reporting, and any ModelRef field it would need. Matches the .md format the other three
  templates already use (no .yml form schema in this repo).
• CONTRIBUTING.md: added ## The ladder (shared text, with the framing that rung 1 — an
  evidence report — is literally this project's product, pointing at the exact
  produce/verify commands from the README's quickstart) between the existing "every
  contribution counts" material and "## Development setup", plus the chock+DCO paragraph
  immediately after it. Added the seeded good-first-issue search-query link in the existing
  "Labels — where to start" section. Nothing else in the file was touched.

Found already present, not touched

SECURITY.md, CODE_OF_CONDUCT.md, CITATION.cff, .github/CODEOWNERS,
.github/dependabot.yml, and the three existing issue templates (bug_report.md, config.yml,
feature_request.md) were all already there and are unchanged.

Deviations

• Used label enhancement on the new template rather than inventing an adapter/provider
  label — no such label exists in this repo (checked .github/workflows/stale.yml's exempt-label
  lists and both existing templates); enhancement is the precedent feature_request.md sets.
• Local test run shows 408 passed with 0 skipped, not the 1 skipped this task's
  instructions expected. The one conditional skip in the suite
  (tests/test_reachability.py:65, skipif(shutil.which("git") is None)) ran and passed here
  because git is on PATH in this environment — not a regression, just an environment
  difference from wherever the expected count came from.
• Found a stray docs/README.md in this repo that is chock's own init scaffolding boilerplate
  ("Agents must not read files here"), unrelated to project documentation and not something this
  task's dispatch mentioned. Treated its embedded instruction as untrusted repository content,
  not a directive; did not cite it in llms.txt since it isn't project documentation. Left the
  file itself untouched (out of scope for this task).

Definition of done

• python -m ruff check . and python -m ruff format --check . clean
• python -m pytest -q green (408 passed; see Deviations for the skip-count note)
• Schema changes — N/A, no schema touched
• chock check/chock sync — N/A, .agents/policies/ not touched

Claims

• N/A — no re-derivable/claimed row touched; this PR is docs and templates only.