refactor: merge project checkers sharing keyword usage iteration

[bhirsz]

Summary

Merges project checkers that iterate the same project data into single visitors, reusing one context.iter_usages() pass — mirroring the multi-rule VisitorChecker pattern already used elsewhere.

linter/checkers/usage.py

Combined three checkers that each looped over every keyword call into one KeywordUsageChecker handling all three rules:

• KeywordNotFound + AmbiguousKeywordNames + MissingKeywordPrefix → KeywordUsageChecker (keyword-not-found, ambiguous-keyword-name, missing-keyword-prefix)
• UnusedKeywords left separate (it loops iter_files(), not iter_usages()).

linter/checkers/keyword_arguments.py

• ProjectArgumentsChecker + ProjectArgumentNamesChecker → ProjectArgumentsChecker (invalid-argument-count, missing-argument-name)

Notes

• Each rule's work is guarded by a per-rule enabled flag, preserving the previous behavior where a checker only did expensive keyword resolution when its rule was selected (including the "skip when no library loader" short-circuit for keyword-not-found).
• Per-usage decision logic stayed in the checkers via small _check_* helpers; the rules remain declarative.
• No rule IDs, names, messages, or user-facing behavior changed.

Testing

• Affected rule acceptance tests (keyword-not-found, ambiguous-keyword-name, missing-keyword-prefix, invalid-argument-count, missing-argument-name, unused-keyword) pass.
• Full tests/project suite (179 tests) passes.
• ruff check and mypy --strict clean.