[docdb] WAL replay of deleted table from colocated tablet can lead to crashes

[bmatican]

Noticed a crash recently that seems to point back to this situation. We should probably have a test for this to validate it indeed can cause this:

• colocated tablet with a table foo
• ALTER TABLE foo -- this requires the table id for foo to be in the TS raft metadata
• DROP TABLE foo -- this will remove the table id for foo from the TS raft metadata
• restart a TS -- on startup, we do a local bootstrap of the rocksdb data + raft data
• if the last rocksdb flush happened before the ALTER, then we need to replay WAL files which would include the ALTER operation, but for a table id that is no longer present in the raft metadata, leading to a crash!