[Durability] Parallel deletes return invalid durability_response 'KeyNotFoundException' [ETA:2019-07-24]

Build: 6.5.0-3748

Scenario:

1. 2 node cluster, Couchbase bucket (replica=1)
2. Load 10 docs initially with durability=MAJORITY (succeeds)
3. Stop Memcached on node-2 using `kill -STOP [pid]`
4. Delete 5 keys (test_docs-1, test_docs-2, test_docs-4, test_docs-7, test_docs-8) from Sdk client-1 with timeout=60, durability=MAJORITY. Client-1 will be blocked until timeout since replica is stopped in step#3
5. From Client-2, try to delete the same keys with durability=MAJORITY

Observation:

Seeing 'com.couchbase.client.core.error.KeyNotFoundException' while trying to delete from client-2. Using the same client-2, when tried to read the values fetches the old values properly.

Expected behavior:

Client-2's delete request should be served with 'com.couchbase.client.core.error.DurableWriteInProgressException'

Attaching pcap and logs for reference.

TAF Testcase:

epengine.durability_failures.DurabilityFailureTests.test_sync_write_in_progress,nodes_init=2,replicas=1,num_items=10,sdk_timeout=60,doc_ops=delete;delete,simulate_error=stop_memcached,durability=MAJORITY