Otherwise we might end up with a search term like '"some junk"*' which is
invalid.

Instead of using distinct alone, change to add the relevance of related
items when searching for tickets. In other words, when performing a full
text search for 'xyz', add the relevance of matched user, organization, and
thread objects.

As currently written, if a full text query is attempted in BOOLEAN mode and
the query cannot be compiled by MySQL, it causes an OrmException and a
crash. This patch adds a more straight-forward and comprehensive
grammer-based approach to checking for valid boolean syntax.

... and some of the related models are null.

Disable sorting on ThreadEntry::email_info__mid lookups for email processing
which results in an on-disk temporary table.

Load and cache all the attachments on a thread prior to rendering the items.
That removes the need to run a query per thread item to check for- and fetch
attachments.

This patch changes the automatic GROUP BY clause for annotations in the ORM
including aggregate queries (like COUNT), by allowing the query to specify
"distinct" fields. If distinct fields are specified, then no other fields
are automatically added to the GROUP BY clause of the compiled query.

This will likely only be supported by MySQL, as other engines and the ANSI
standard require all non-aggregate fields in the SELECT clause to be present
in the GROUP BY clause.

If the configuration for the file backend is set to something invalid, do a
better job of falling back to database-backed attachments.

This feature allows (by option) an organization to allow its members to see
tickets from other members. It can be used beside or instead of the
automated collaboration feature. Now, collaborators retain access to tickets
on which they participate; however, sharing can be used to enable access to
other tickets owned by their organization.

include @greezybacon suggestion on fix

Fixes #2350 in `develop`