Fix CSV sniffing samples truncated in the middle of a line

[Voyagerroc-Lab]

Piping a valid TSV with long rows into rich - --csv can exit with Could not determine delimiter: the 1,024-character sniffing sample ends partway through a row. This also affects comma, semicolon, and pipe delimiters.

Sample the first 100 complete physical lines for both delimiter and header detection. This bounds the number of sampled lines while allowing fields longer than 1,024 characters. Rendering still consumes the full input, including rows beyond the sample and the existing head/tail selection.

Related to #54. This addresses truncated sampling; csv.Sniffer remains heuristic. #89 provides a separate manual dialect override and retains the truncated sample in its automatic path. The extension fallback from #39 remains covered by tests.

Validation on Windows, Python 3.11.9, in a fresh virtual environment (pip install -e . pytest click==8.1.8 black==22.3.0; Click 8.1 retains the FC alias used by the existing test):

• Before the fix: new tests had 13 failures, 2 passes, with delimiter-detection errors; the existing test passed separately.
• After the fix: python -m pytest -q — 16 passed.
• Covers all four supported delimiters, 300- and 1,500-character fields, stdin with LF/CRLF, output beyond the sample, missing final newline, head/tail, and CSV/TSV extension fallback.
• Black 22.3.0 checks on both changed files, pip check, and git diff --check pass.