sqlccl: make IMPORT CSV tolerant of data errors

[knz]

Reported in #25480

If any line in the CSV file contains erroneous (unparsable/incorrect data), the entire IMPORT job fails. This is poor UX, because the amount of work needed to correct the data might be negligible compared to the time needed to complete the IMPORT job.

Each of the following alternatives would be better UX, and the combination of them even better:

• make IMPORT save erroreous CSV lines in a separate file (possibly also with location information) for manual inspection, and skip over them
• make IMPORT suspend on data error with a status message in the job table that indicates where the error was found, to let the user fix the error and then resume the IMPORT job
• make IMPORT accept SQL options to specify what to do in case of data errors, including skipping over erroneous lines

Jira issue: CRDB-5695

[msbutler]

triaging to sql-exp as new owners of this.

Ideally ,the log_ignored_statements option should no longer be undocumented and experimental for IMPORT INTO.