[Python] Extend csv `invalid_row_handler` with `fill_null` and `ignore_unknown`

Referencing a past comment by @jorisvandenbossche,

_A bit orthogonal to this PR, but while looking at it: it could be an extension to allow an additional handler return value of "fill_null" (or another name), which could signal that in case the row has fewer fields, the missing fields can be filled with nulls (this is actually the default behaviour in pandas)._

_That's at least theoretically doable, perhaps you can open a separate JIRA for that?_

_Originally posted by @pitrou in https://github.com/apache/arrow/issues/12012#issuecomment-1002637825_

Along with adding `fill_null` value, it would be good "the opposite", `ignore_unknown` (or `ignore_value`) in case there are more unpacked values than columns in the schema.

Start by locating the Python CSV invalid_row_handler entry point and its existing tests. Read how handler return values are interpreted for rows with missing or extra fields. Done means both fill_null and ignore_unknown are supported and covered by tests for the respective row-shape cases.

Written by the indexing model from the issue text.