Validate scraper config (warn on unknown keys, reject bad types)

[cybermejo]

Summary

Adds config validation to the scraper so typos and wrong types are caught up front instead of silently ignored or crashing mid-run.

Why

load_config did {**DEFAULTS, **cfg} with no checks. A typo like searchTerm (missing s) was silently dropped, and a wrong type such as "maxItems": "100" only failed deep inside the run after the config looked fine.

What changed

• New validate_config():
  • Unknown keys → warn (not error). The config schema mirrors the Apify Tweet Scraper V2 input, which is a superset of what this script implements, so a real Apify vars.json must still load unchanged. Warning surfaces likely typos without breaking compatibility.
  • Known keys with the wrong type → hard error (sys.exit) reporting the key, expected type, and what was received.
  • A non-object top-level config is rejected up front.
• _CONFIG_TYPES maps each known key to its accepted type(s) (type(None) marks nullable fields).

Behavior (smoke-tested)

example config            -> runs clean
unknown keys (Apify cfg)  -> "[warn] unknown config key 'addUserInfo' — ignored." then runs
maxItems: "lots"          -> "Config error: 'maxItems' must be int or float, got str." (exit 1)

Testing

Full validation loop green from scraper/ (ruff, ruff format --check, mypy, pytest). Adds tests/test_config.py (9 tests); fully offline.

Independent of #1 — touches load_config, a different region of tweet_scraper.py than the streaming change.