Implement support for applying updates as deletes + inserts

This needs to happen at the higher layer of writing from Spark/Flink etc. 

Hudi can already support this, by 

• Logging delete blocks to the old file group. 
• Writing new data blocks/base files to the new file group.