handle IO errors between store_log_entry and end_of_append_batch

[Besroy]

In the current implementation, after all data is received in a batch, raft saves the log entries into the log store and performs pre-commit. Then the end_of_append_batch step ensures that all data is written. However, if an IO error occurs between the save_log_entry and end_of_append_batch stages, the error may cause the HomeStore to become stuck (https://github.com/eBay/HomeStore/blob/master/src/lib/replication/repl_dev/raft_repl_dev.cpp#L454) or crash(https://github.com/eBay/HomeStore/blob/master/src/lib/replication/repl_dev/raft_repl_dev.cpp#L873).
For example:
t1: handle_raft_event and pass ( all data received)
t2: append log to log store and add rreq into state machine
t3: precommit pass
t4: async write failed at on_push_data_received, trigger handle_error / on_fetch_data_received and crash
t5: end_of_append_batch, wait for all data written, stuck or crash

Note that we cannot do unlink at handle_error if IO error occurs directly(in the following case), so we need to find a solutaion to handle error more gracefully, and one potential approach could be emergent gc.
t1: Handle the raft event and pass.
t2: Append log to the log store and save the LSN into the state machine.
t3: Fail to write due to an IO error, and then trigger handle error and remove it from the state machine.
t4: In the pre-commit phase, can't find the LSN, leading to a nullptr exception.

Also need flip to mock IO error.