Some data may be lost if a table is flashed back

[JaySon-Huang]

https://github.com/pingcap/tics/blob/33678430328cfee1b63750841419a3b70a4d528c/dbms/src/Storages/Transaction/PartitionStreams.cpp#L33-L41
Was introduced by #599.

The actions between applying raft log and schema changes are not strictly synchronized. There could be a chance that some raft logs come after a table gets tombstoned. If we drop those data once a table gets tombstoned, then some data may be lost after that table is flashed back later. @zanmato1984

[ilovesoup]

What is the reason for downgrade priority? @zanmato1984

[ilovesoup]

Before fix, we need to label this as limitation of tiflash and give error when doing flash back (instead of reading)