Memtable support dedup rows with same primary keys

[jiacai2050]

Describe This Problem

Current memtable implementation will always append row to inner list, which will take more memory than expected when there are duplicated rows, and this will cause SST in level0 is very small, which is unfriendly for query.

Proposal

Dedup rows when insert to memtable for tables in overwritten mode

Additional Context

https://github.com/CeresDB/ceresdb/issues/1020#issuecomment-1608889115