Validate multiLine reads against runaway quote consumption

[Iwan-Dyke]

Problem

With multiLine=true (merged in #17), a malformed file with unmatched quotes causes Spark to silently swallow multiple lines into a single field. The result is fewer rows with bloated field values — no error raised, _corrupt_record not triggered.

This is the inverse of the #16 bug: previously corruption showed as extra rows, now it shows as missing rows with oversized fields.

Suggested approach

• After read, compare actual row count against a raw newline count heuristic
• Or flag any field value exceeding a configurable max length
• Or both, behind a validate_multiline=True option

Acceptance criteria

• A file with unmatched quotes produces a warning or error, not silent corruption
• Validation does not break well-formed multiLine files
• Behaviour is documented