…ightening

- New filters per server FILTERS_MAPPING: datatype, creator, and
  sentiment_score across latest_api, archive_api, market_api, count_api,
  and market_count_api. Plus filling in existing-bucket coverage gaps in
  archive/count (removeduplicate, sentiment, tag, region, organization).
- New _FLOAT_PARAMS frozenset and _check_float_param helper for
  sentiment_score (numeric; rejects bool via int-subclass check, accepts
  int alongside float).
- Count endpoints now support scroll=True. _scroll_all carries the
  is_count flag through _dispatch and captures the final aggregate dict
  under the merged response's "aggregate" key when the API returns
  results as a dict on the final page.
- Client-side enforcement of server's HTTP 422 cases:
  - size capped at 50 (paid-plan maximum); values > 50 raise
    NewsdataValidationError pre-flight.
  - 5 mutex groups mirrored from the server validator: q / qInTitle /
    qInMeta; country / excludecountry; category / excludecategory;
    language / excludelanguage; and the 3-way domain / domainurl /
    excludedomain group.
- One-line docstring added to _dispatch for parity with the other
  internals.
- 23 new tests covering all the above; unit suite 99 -> 122. All checks
  pass on Python 3.12 and 3.14.

- size: enforce lower bound (must be in [1, 50])
- raw_query: reject keys with empty values
- sentiment_score requires sentiment