Detect HoodieMerge type with engine type automatically instead of the default avro-based one

The same table be read by Spark or Hive for e.g I would love for such queries to be able to read and merge using the engine' POJO (Spark -> Row, Hive -> ArrayWritable), if the user chooses to write such a "merge" implementation. 

For e.g consider this API
{code:java}
Option combineAndGetUpdateValue(HoodieRecord older, HoodieRecord newer, Schema schema, Properties props) {
...
} {code}
when writing from Spark we would use `HoodieSparkRecord` and the merge class would be `HoodieSparkMerge` in the table properties? and if we now read with Hive, it falls back to avro somehow? 

## JIRA info

- Link: https://issues.apache.org/jira/browse/HUDI-4301
- Type: Improvement

Start by tracing the HoodieRecord combineAndGetUpdateValue API and the table properties that select the merge implementation. Compare the Spark and Hive read/write paths, then define how the engine-specific merge type is detected and how Avro remains the fallback; done means both engines can read and merge compatible tables without manual default selection.

Written by the indexing model from the issue text.