KAFKA-10561: Preserve sub-millisecond precision in TimestampConverter

[michal-swiatowy]

NOTE: I used https://issues.apache.org/jira/browse/KAFKA-10561 for my PR but not sure how much it matches, I'm happy to change or create a new issue if required. Maybe https://issues.apache.org/jira/browse/KAFKA-10627?

Main idea
Replace java.util.Date (millisecond precision) with java.time.Instant (nanosecond precision) as the internal representation in TimestampConverter. This fixes loss of microsecond and nanosecond data when converting between unix and string timestamp formats.

To do this a new format.type configuration option is added that selects the string parsing/formatting engine:

• SimpleDateFormat (default) - for backward compatibility,
• DateTimeFormatter - uses java.time.format.DateTimeFormatter

Key invariants

• Default behavior is unchanged - SimpleDateFormat remains the default format.type
• Conversions involving Date, Time, or Timestamp Connect logical types still go through java.util.Date at the boundary and remain limited to millisecond precision (this is a Connect API constraint).
• Using only JDK java.time classes.

Testing

• Added new tests for data format support and various convert cases
• Full connect:transforms test suite ran ok

Sample usage

transforms.ts.type=org.apache.kafka.connect.transforms.TimestampConverter$Value
transforms.ts.target.type=unix
transforms.ts.unix.precision=microseconds
transforms.ts.format=yyyy-MM-dd'T'HH:mm:ss.SSSSSS'Z'
transforms.ts.format.type=DateTimeFormatter