fix(csv): detect the dialect when a quoted field spans several lines

[r0h1tb]

Problem

A CSV whose first field is quoted and contains a newline fails to convert:

DocumentConverter(allowed_formats=[InputFormat.CSV]).convert(
    DocumentStream(name="quoted.csv",
                   stream=BytesIO(b'"line one\nstill line one";b;c\n1;2;3\n')),
    raises_on_error=True,
)
# RuntimeError: Pipeline SimplePipeline failed
#   _csv.Error: ',' expected after '"'

convert() sniffs the dialect from self.content.readline(). That read stops at the newline inside the quoted field, so the sniffer gets '"line one' — an unterminated quote — and raises csv.Error. The handler falls back to csv.excel, and reading a semicolon file as comma-delimited under strict=True aborts the conversion.

Fix

Retry the sniff on a larger sample when the first line fails; the sample closes the quote and the dialect is detected.

The first line is still tried first, deliberately. Sniffing the larger sample unconditionally regresses existing fixtures: csv.Sniffer rejects samples whose rows carry different numbers of delimiters (csv-inconsistent-header, csv-too-many-columns), and it also infers quotechar from what it is given — on csv-too-few-columns it picks ' and strips the quotes from a 'b' cell, changing that file's groundtruth. Falling back only on failure leaves every currently-detected dialect untouched.

Tests

test_quoted_newline_in_first_field in tests/test_backend_csv.py. Against unfixed code:

FAILED tests/test_backend_csv.py::test_quoted_newline_in_first_field
  - RuntimeError: Pipeline SimplePipeline failed

tests/test_backend_csv.py goes 3 passed → 4 passed; no groundtruth was regenerated. ruff check and ruff format --check are clean.

Found while reading the backend, so there is no tracking issue. Related but distinct: #1716 is the single-column sniff failure already handled by the comma fallback.

[github-actions]

✅ DCO Check Passed

Thanks @r0h1tb, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🟢 Merge protection satisfied — ready to merge.

Show 1 satisfied protection

🟢 Enforce conventional commit

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?:

[codecov]

Codecov Report

❌ Patch coverage is 92.85714% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
docling/backend/csv_backend.py	92.85%	0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

[ceberam]

Thanks @r0h1tb for spotting this issue and suggesting this PR.
The fix looks correct to me. Just a couple of comments worth addressing.

[r0h1tb]

Both points addressed in a4f6ba9: sample capped at 4096, and it is now only read when the first-line sniff raises. _sniff_dialect takes a callable so the fallback read stays on the fallback path. CSV suite 13 passed; ruff check and format clean.

[ceberam]

Thanks @r0h1tb for addressing the comments.