* Go back to time window relative to oldest fragment -- but only build
1/4 of the window to avoid 'seam' problems with the next block of data.
The assumption is that a next block should come from a prior source within
3/4 of the build window.
* Fix a problem with initializing the oldest received time in findOldest()
that was making input queues get totally flushed...since the oldest time
chosen was always ~0. Problem is that it appears in some contexts the
compiler is treating time_t (unsigned long according to the headers) as
a signed value so nothing positive is < it.