[ngrams] ignore errors when parsing TSV files

Related to
#433

Textual formats often have either encoding or column errors.

This is a common issue, see e.g.
https://duckdb.org/docs/data/csv/reading_faulty_csv_files.html

We just ignore them.