GET_REPLICA for a key which has a prepared SyncWrite against it incorrectly returns SyncWriteReCommitInProgress

Steps to reproduce:

1. Create a 2 node cluster and default bucket with replica=1
2. Start inserting documents using java sdk in 1 thread and using another thread start fetching(getfromReplica) the last inserted doc using the same client connection. In the test this is done for 100 items and out of which the durability failed for the below mentioned documents.
Note: Here, read thread is running 1 behind the write thread so that it will read the last inserted document

Actual Result:
getfromReplica should return the document from both active and replica vBucket but for the mentioned documents getfromReplica contains result from 1 node which ideally should be active.

Example:
Key = test_docs-7 getfromReplica = [{'cas': 1563419755079401472L, 'value': {}, 'key': 'test_docs-7', 'status': True}]

Expected Result:
If getfromReplica is at all fetching the result it should contains 2 results from both nodes.

Example:
Key = test_docs-0 getfromReplica = [{'cas': 1563419754994991104L, 'value': {}, 'key': 'test_docs-0', 'status': True}, {'cas': 1563419754994991104L, 'value': {}, 'key': 'test_docs-0', 'status': True}]

Out of 100 docs, Durability failed for:
['test_docs-14', 'test_docs-51', 'test_docs-57', 'test_docs-32', 'test_docs-11', 'test_docs-55', 'test_docs-49', 'test_docs-47', 'test_docs-86', 'test_docs-67', 'test_docs-7']