fix(push): set LazyQuotes on CSV readers (completes #222 — avoids a false-reject)

[saadqbal]

Why this is a follow-up to #222

#222 landed the fail-closed-on-read-error change, but was merged before the root-cause commit could be added to it — so develop currently has a regression: openCSVReader sets FieldsPerRecord=-1 but not LazyQuotes, so a bare/unescaped quote (which pandas/the ingestor tolerates) makes csv.Reader.Read return a parse error. Combined with #222's fail-closed change, a bare-quote CSV now hard-fails preflight — a false-reject of a layout the cluster ingests cleanly. (Before #222 it was a silent skip / fail-open; #222 correctly closed that, but without LazyQuotes it swung to the opposite wrong.)

Fix

Set LazyQuotes = true centrally in openCSVReader (fixes CrossCheckLabels, readLabelColumnValues, CheckSequenceRows, CheckHasDataRows, ReadCSVHeader at once) and in InferSchema's own reader. The readers are now pandas-faithful; the remaining non-EOF Read error is genuinely I/O-only, so #222's fail-closed handling is correct. Added a bare-quote guard test; the full value/verdict parity harness passes, confirming LazyQuotes doesn't over-accept relative to the ingestor.

Scope

Completes the read-error fix set (#221/#222) surfaced by Bugbot on #219; rolls up under release ticket #220; should land before v0.8.0 so the release doesn't ship the false-reject.

🤖 Generated with Claude Code

---

Note

Medium Risk
Touches all pandas-mirrored preflight CSV walks and schema inference; wrong quote handling could still false-reject or over-accept vs the ingestor, though parity tests are meant to guard that.

Overview
Turns on LazyQuotes on the shared openCSVReader path and on InferSchema’s reader so bare/unescaped quotes in data rows are tolerated like pandas/the ingestor, instead of making Go’s strict csv.Reader fail preflight.

That closes the regression from fail-closed read handling (#222): CSVs the cluster would ingest were being false-rejected at preflight when a row contained a quote pandas keeps. With FieldsPerRecord = -1 and LazyQuotes, non-EOF read errors are treated as real I/O failures only, so existing fail-closed behavior stays correct.

Adds TestOpenCSVReader_LazyQuotesMatchesPandas so label diversity and value reads still see all rows (e.g. three rows, two classes) on a bare-quote fixture.

^{Reviewed by Cursor Bugbot for commit 9262748. Bugbot is set up for automated code reviews on this repo. Configure here.}