Preserve missing CSV fields with custom null regex

[osipovartem]

Summary

• Preserve the number of fields actually present in truncated CSV rows so absent fields remain SQL NULL even when a custom null matcher preserves empty strings. Strict reads allocate no field-count buffer.
• Add opt-in quoted-empty tracking: "" can remain an empty string while an unquoted empty field maps to NULL. The raw-record scanner and flags are absent from the default path. Schema inference with this mode and an empty null marker requires an explicit schema; otherwise it returns a clear error.
• Add exact null-value matching with a small-list linear path and a hash set for larger lists. Regex matching remains available for non-literal patterns.

Validation

• cargo +1.97.0 test --offline -p arrow-csv --lib: 85/85 passed, including chunk boundaries, blank lines, comments, CRLF, custom terminators, truncated rows, escape/quote precedence, and NUL quote.
• cargo +1.97.0 clippy --offline -p arrow-csv --all-targets -- -D warnings -A clippy::for_kv_map: passed; the suppression is for an unrelated existing warning in arrow-array.
• cargo +1.97.0 fmt --all -- --check and git diff --check: passed.
• Local release benchmark, 1M CSV rows, 5 measured iterations: default 47.48 Mrows/s; regex NULL_IF 34.05; exact values 40.70; quoted-empty exact values 27.43. Quoted tracking is opt-in and does not affect the default path.
• Independent read-only review: approved after addressing the reported parser edge cases.