Configurable error handling strategy when target partition does not exist during Stream Load

[liujunfei2230]

When using the Flink Doris Connector to write data to partitioned tables via Stream Load, if the target partition does not exist, the connector currently only has one behavior: retry and then throw a DorisBatchLoadException, which causes the entire Flink job to fail.

This is problematic in real-world streaming scenarios where:

Historical or dirty data may arrive with partition values outside the expected range (e.g., a record from last year hitting a table that only retains 30 days of partitions), causing the entire job to crash.
In multi-table sync scenarios, a single table's partition issue can bring down the entire pipeline, affecting all other tables.

Add a configurable error handling strategy for "partition does not exist" scenarios. Suggested options:

Strategy	Behavior
FAIL (default)	Throw exception, job fails — current behavior, backward compatible
SKIP	Silently skip the record, continue processing
SKIP_WITH_WARN	Skip the record but log a warning with the table name and partition value
AUTO_CREATE	Automatically create the missing partition via ALTER TABLE ADD PARTITION before retrying the write