Split tweet_scraper.py into a config/query/api/cli package

[cybermejo]

Summary

Splits the 630-line tweet_scraper.py into a config / query / api / cli package — the last remaining README Roadmap item for the scraper. Behavior-preserving: same 65 tests pass.

Stacked on #9 (base worktree-scraper-package-11, the logging+packaging PR). GitHub will retarget this to main once #9 merges.

Layout (flat public API preserved)

from tweet_scraper import X and the tweet_scraper:main entry point are unchanged — __init__.py re-exports the full public API (including the _underscore helpers the tests import).

Module	Contents
config.py	DEFAULTS, _CONFIG_TYPES, validate_config, load_config, date parsing
query.py	build_query, handle_from_url
api.py	endpoint URLs, get_api_key, _extract_*, paged_get (retry/cursor), search_tweets, user_tweets, tweet_replies
cli.py	CSV_COLUMNS, flatten, streaming writers, main
__main__.py	enables python -m tweet_scraper

Mock-patch retargeting (the tricky part)

The suite has 20 patch targets on the flat namespace. Since the patched objects moved into api, the strings were retargeted to where names are actually looked up — tweet_scraper.api.{requests,time,paged_get,tweet_replies} — and cli calls fetchers via the api. namespace so the patches intercept.

Verification (beyond the green loop)

• Falsification test: deliberately pointing one patch at a wrong target makes the tests fail, and restoring it makes them pass — proving the mocks genuinely intercept rather than passing inertly.
• 3-agent adversarial review: confirmed byte-for-byte logic equivalence vs. the original, complete public-API re-export, and an acyclic import graph with correct patch indirection. Two minor findings it raised (missing python -m entry, stale docs) are fixed in this PR.

Also updated

• CI mypy tweet_scraper.py → mypy tweet_scraper.
• README + CLAUDE.md: invocation → python -m tweet_scraper, project-structure tree, data-flow, and Roadmap (shipped items marked done).

Testing

Full CI-equivalent loop green from scraper/: ruff, ruff format --check, mypy tweet_scraper (6 files), pytest --cov — 65 tests pass. python -m tweet_scraper --dry-run verified without install.