[Java] Inconsistent logic for type IDs in Union vectors

The current logic for calculating the type IDs in UnionVector#getField and DenseUnionVector#getField is:
1. DenseUnionVector uses an increasing counter 
1. UnionVector uses the ordinal of the type enum
1. Both completely ignore the type IDs provided at construction as part of fieldType (if provided)

We encountered this inconsistency while testing a direct roundtrip of a union vector between pyarrow and Java with the C Data Interface ('direct' here means without using VectorSchemaRoot/RecordBatch). The identifiers for the type IDs differ after completing a roundtrip. 

 

**Reporter**: [Roee Shlomo](https://issues.apache.org/jira/browse/ARROW-14271) / @roee88

_{**Note**: *This issue was originally created as [ARROW-14271](https://issues.apache.org/jira/browse/ARROW-14271). Please see the [migration documentation](https://github.com/apache/arrow/issues/14542) for further details.*}

Start with UnionVector#getField and DenseUnionVector#getField, then trace how fieldType and its provided type IDs are handled during construction. Reproduce the direct pyarrow-to-Java C Data Interface roundtrip and verify that both vector types preserve consistent type IDs after the roundtrip.

Written by the indexing model from the issue text.