Add CI lint workflow and fix all ruff lint + format errors

[austrian-code-wizard]

Summary

This PR adds a GitHub Actions CI workflow (.github/workflows/lint.yml) that runs ruff check and ruff format --check on PRs and pushes to main. It also fixes all pre-existing ruff lint errors and applies ruff format across the entire codebase so CI passes cleanly.

Lint fixes applied:

• F401 (unused imports): Removed unused imports in setup.py, notebooks/collie_feedback_gen.ipynb, src/feedback/gpt_content.py, src/feedback/gpt_style.py, src/feedback/manual.py, src/lcdpo.py, src/train.py, tests/test_metrics.py
• E731 (lambda assignments): Converted metric = lambda ... to def metric(...) in src/eval.py and src/train.py
• F541 (f-strings without placeholders): Removed extraneous f prefixes in src/modal/app.py and src/sample.py
• E402 (import ordering): Moved np.random.seed(42) after all imports in src/sample.py
• E701 (multiple statements on one line): Split inline else: in src/train.py
• E712 (boolean comparisons): Replaced == True / == False with idiomatic assertions in tests/test_metrics.py

Formatting applied (ruff format):

• Applied ruff format to all 29 files that were not conformant — line wrapping, quote normalization, trailing commas, whitespace cleanup, etc. These are purely cosmetic changes with no semantic impact.

The deliberate unused imports in src/logger.py (originally added to test CI failure detection) have been removed as part of the fixes.

Review & Testing Checklist for Human

• Verify the lambda-to-def conversions in src/eval.py and src/train.py preserve correct closure/scoping behavior (the metric function references feedback.metric and feedback.metric_value from the enclosing scope)
• Confirm np.random.seed(42) being moved after all imports in src/sample.py doesn't change execution order in any import-time side-effect scenario
• Check that removed imports (langdetect.detect, re in tests, EOS_EMOJI_REGEX in manual.py, Metric in gpt_content/gpt_style, Optional in lcdpo, dump_arg_dicts in train, find_packages in setup) are truly unused
• Spot-check a few of the ruff format changes in the larger files (e.g. src/feedback/collie.py, src/eval.py) to confirm no unintended semantic changes crept in through line-wrapping or string reformatting

Test plan: CI lint job passing is the primary validation. No runtime behavior should change — all fixes are cosmetic (unused import removal, style fixes, boolean comparison idioms, formatting).

Notes

• The workflow runs on Python 3.11 with the latest ruff version
• Both ruff check and ruff format --check now pass in CI

Link to Devin session: https://staging.itsdev.in/sessions/1ce0ad01211c45779eeb3d574e6ba972
Requested by: @austrian-code-wizard