This reverts commit 3c45701.

Currently, AppendService will loop forever waiting for an appended
journal to spring (back) into existence if it's not found.

Instead, immediately fail the AsyncAppend and all of its dependents.

This is desirable because journal deletions are often asynchronous and
race deployed shards, and there's not much that can be done aside from
failing and handling the error at a higher level.

This happens when a journal is new and the topology is converging.
It shouldn't result in a terminal Append failure.

It's not uncommon for recovered ACKs to contain journals which have
since been deleted. When this happens, log a warning and otherwise
discard the ACK intent.

Do this only for recovered intents: writen ACKs of the current session
must still exist.