Add main-based PR evaluation harness

[yashturkar]

Summary

• add a harness for reviewing a target ref from an isolated -based worktree
• document the harness in a dedicated runbook and the repo README
• restore the required docs directory so docs lint passes on

Verification

• python3 -m py_compile scripts/eval_pr.py
• python3 scripts/eval_pr.py --help
• kb-server/.venv/bin/python scripts/docs_lint.py
• python3 scripts/eval_pr.py HEAD --tests-only --tmux-session-name fd-pr-eval-head-2
• python3 scripts/eval_pr.py HEAD --e2e-only --tmux-session-name fd-pr-eval-e2e-2

[yashturkar]

Addressed the Inspector harness blockers:

• resolve symbolic refs like HEAD to a commit SHA before checkout
• track harness-owned worktree/tmux resources so pre-existing sessions are left alone and harness-created artifacts are preserved on failure or --keep-temp
• add focused unit tests and doc updates for the new behavior

Verification run in /tmp/flight-deck-pr13:

• python3 scripts/docs_lint.py
• python3 -m py_compile scripts/eval_pr.py tests/test_eval_pr.py
• python3 -m unittest tests/test_eval_pr.py

[yashturkar]

Fixed the generated docs check by switching generation dates to UTC in Generated docs/generated/api-surface.md and docs/generated/env-catalog.md and refreshing plus .

[yashturkar]

Fixed the generated docs check by switching generation dates to UTC in scripts/generate_context_artifacts.py and refreshing docs/generated/api-surface.md plus docs/generated/env-catalog.md.