Use `tokenize` for linter benchmark (#11417)
## Summary
This PR updates the linter benchmark to use the `tokenize` function
instead of the lexer.
The linter expects the token list to be up to and including the first
error which is what the `ruff_python_parser::tokenize` function returns.
This was not a problem before because the benchmarks only uses valid
Python code.
