refactor: enforce consistent lint and formatting with pre-commit and CI

[LiberiFatali]

Why

The recent CI failure was caused by small lint and formatting issues that were only caught after pushing. This PR makes those checks happen earlier — on the developer's machine before commit — and re-enables them as a required CI check, so the same failure won't happen again.

What this PR does

• One-time cleanup: Formats the whole codebase with a single tool so all files are now consistent. Future blame views ignore this commit.
• One formatter: Removes the duplicate black setup and keeps ruff as the single source of truth for both linting and formatting.
• Local guard: Adds a pre-commit hook (ruff check + ruff format) that auto-fixes simple issues on every commit.
• Remote guard: Restores the formatting check in CI on changed files — it can't be bypassed, even if the hook is skipped locally.
• Guide: Adds CONTRIBUTING.md with setup and usage instructions.

How to use after this lands

uv sync --frozen
uv run pre-commit install   # run once
# hooks run automatically on commit; run manually with:
uv run pre-commit run --all-files

Hooks fix what they can locally; CI is the final gate. mypy remains available locally (uv run mypy) but is not enforced.

[LiberiFatali]

@mambo-wang Many of the code changes involve only reformatting. I have confirmed that all tests passed.

[mambo-wang]

Reviewed and verified locally against the merge base (e92692b1):

• Blast-radius check: AST-compared all 184 changed .py files. Differences are limited to quote normalization, docstring whitespace normalization, signature re-wrapping, and F401 import removals. Every removed import was cross-checked repo-wide for re-export/consumer usage — none found (e.g. ComponentMeta consumers already import from codewiki.mcp.cache; validate_api_key consumers import from utils.validation directly).
• Runtime: full test suite passes (428 passed / 2 skipped; the single failure in test_review_changes.py::test_prepare_end_to_end reproduces on the develop baseline and is pre-existing). Repo-wide ruff check . and ruff format --check . are green.
• MCP surface: tools/list (44 tools), prompts/list (18) and resources/list (3) are byte-identical between base and head; live tools/call smoke tests succeed.

Two non-blocking follow-ups after merge: (1) append the actual squash-merge commit hash to .git-blame-ignore-revs (currently placeholder-only, and the suggested --grep string does not match this commit's message); (2) heads-up that the PR also reformats embedded Python code blocks in 24 markdown files — harmless, though generated repowiki/ docs will get the old formatting back on regeneration.