fix: handle UnicodeDecodeError in CsvConverter when charset detection is inaccurate (fixes #1949)

[hanhan761]

Summary

CsvConverter.decode() uses stream_info.charset (detected from partial file content) to decode CSV files. When the detected charset is inaccurate (e.g., 'ascii' from the first 4096 bytes, but the full file contains UTF-8 characters), the decode fails with UnicodeDecodeError.

This fix adds a fallback: if decoding with the detected charset fails, use charset_normalizer for automatic charset detection on the full file content.

This is the same class of bug as #1505 (PlainTextConverter), which was fixed in PR #1938.

Issue

Fixes #1949

Verification

• UTF-8 CSV files with non-ASCII characters now decode correctly
• Pure ASCII CSV files continue to work as before