First part of efficiency fixes for message sending to pid's (cutting down

First part of efficiency fixes for message sending to pid's (cutting down
the amount of time we hold tdb locks). Gulp down all messages at once rather
than reading/re-writing one at a time. NOTE: All dispatch routines *must*
be able to cope with incoming message on *odd* byte boundaries (all current
handlers do).
Jeremy.
(This used to be commit b752c0340f96669b2b2573cf7d3d10f99580b070)