Introduce Johnsoning optimization process for Lucene indexes

A big part of making index searches fast in Jira is ensuring we have the indexes divided into as little segments as possible (less fragmentation), and currently we only put all documents into a single segment in a "Lock Jira and rebuild indexes" process through an optimize() call.

We should have an option for Jira administrators to call optimize() apart from a foreground reindex, since this should be a faster approach with the same end-result. This way it is possible to run a background reindex and optimize later.

The optimize() call, however, should still lock Jira (using a Johnson page?) as this holds the write lock of indexes for an extended period of time and could cause searches or updates to fail.

We should also inform the administrator that this process will not correct corrupted indexes - perhaps not even allow its use should we detect corruption like we do with background reindex.

Note on fix

Jira 8.0 has new Lucene 7 and doesn't need optimize() as it is done automatically.