GC delete range doesn't work when the cluster has TiFlash

[MyonKeminta]

Bug Report

1. What did you do?

In a cluster that has TiFlash, drop a table (which causes delete range in GC after gc_lifetime expires)

2. What did you expect to see?

GC runs correctly

3. What did you see instead?

GC's delete range phase finishes with error like:

[2020/03/17 14:54:38.206 +00:00] [ERROR] [gc_worker.go:590] ["[gc worker] delete range failed on range"] [uuid=5c31204fb30000d] [startKey=74800000000000005f] [endKey=748000000000000060] [error="[gc worker] destroy range finished with errors: [rpc error: code = Unimplemented desc =  rpc error: code = Unimplemented desc =  rpc error: code = Unimplemented desc = ]"]

4. What version of TiDB are you using? (tidb-server -V or run select tidb_version(); on TiDB)

v4.0.0-beta.1

To fix this, TiDB should skip sending UnsafeDestroyRange to TiFlash stores. It can distinguish it by the store's labels. However in case the labels are not set correctly, it's possible that TiKV node can be skipped too (see tikv/tikv#7153 )