[3.15] gh-110128: Ignore a truncated final row in csv.Sniffer

[Punisheroot]

Fixes #110128.

When a sample is truncated in the middle of its final row, the incomplete row can distort the character-frequency table used by _guess_delimiter() and cause Sniffer.sniff() to raise csv.Error.

This change excludes the final physical row from delimiter detection when the sample contains multiple rows and does not end with \r or \n. A single unterminated row remains usable.

The handling is intentionally kept within the existing _guess_delimiter() row preparation, making this a targeted maintenance-branch fix without changing the delimiter-detection algorithm. It mirrors the cut and len(rows) > 1 behavior of the rewritten Sniffer in Python 3.16. If maintainers would prefer the truncation handling at a different level, I’m happy to adjust the implementation.

Tests cover:

• single-row samples without a trailing newline;
• truncated samples with LF line endings;
• truncated samples with CRLF line endings.

• Issue: CSV dialect sniffer gets wrong dialect (excel) when tab delimited csv file has been truncated #110128