library prefix / field processor cleanup

[David Bremner]

I started trying to generalize the regexp stuff to include
message-ids [1], but didn't get that far.  I think both of these patches
are defensible on their own, particularly the second. I'll probably
include something like them in a revised series for regexp searching.

Probably we should use more of the the "release()" style memory
management with Xapian; I didn't follow that up yet as I didn't want
to get _too_ distracted from current goal.


[1]: the complication is that message-ids are boolean fields, while
     from and subject are probabilistic. It's not insurmountable, but it
     did inspire re-thinking the way query fields are initialized.