Fix import edge cases for DATE/TIMESTAMP, INT96, CSV quotes and format filtering

[gengziyand]

Summary

This PR fixes several import edge cases in java/tools:

• Support DATE and TIMESTAMP value conversion before writing Tablets, and pass import time precision from
  TabletBuilder to ValueConverter.
• Decode legacy Parquet INT96 timestamps from nanos-of-day plus Julian day, and mark them as nanosecond precision in
  auto schema.
• Resolve timezone offsets from the parsed local datetime and configured ZoneId, instead of using the JVM's current
  DST offset.
• Make --format filtering in directory mode respect file extensions, so unrelated files are skipped instead of parsed
  as data.
• Extract Arrow DATE and TIMESTAMP vectors through their actual vector classes.
• Parse quoted CSV fields with embedded separators and escaped quotes.

Tests

mvn '-Dspotless.apply.skip=true' test
'-Dtest=ValueConverterTest,TabletBuilderTest,ParquetSourceReaderTest,DateTimeUtilsTest,TsFileToolFormatFilterTest,CsvSou
rceReaderTest,ArrowSourceReaderTest,TsFileToolEndToEndTest'
mvn '-Dspotless.apply.skip=true' test