fix: coerce out-of-bounds nanosecond timestamps to NaT instead of raising

Dates beyond ~2262-04-11 overflow pandas' int64 nanosecond representation.
When a query returns a timezone-aware datetime column containing such a value
(e.g. a sentinel 'far future' date like 3118-01-01), pd.to_datetime() raises
OutOfBoundsDatetime. In result_set.py the broad except logs at ERROR level,
causing the error to surface repeatedly in observability tooling for any
auto-refreshing chart that queries the affected table.

result_set.py: pass errors="coerce" so out-of-range timestamps become NaT
rather than raising.

commands/dataset/importers/v1/utils.py: extract _convert_temporal_columns()
helper and catch OutOfBoundsDatetime specifically — malformed dates still
raise (original behavior preserved), only out-of-bounds timestamps are
coerced to NaT with a warning count.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>