pc?ul: copy/paste bug in walk_1_set

Values fixed to be squares were tested as primes instead.

This may require some calculations to be redone; however few if any
cases are expected, since in most cases if any values are fixed to be
square we don't do the further allocations that could cause this path
to be reached.

c?ul: disp_batch() should show only allocations for the batch

Track the level this batch required, and show only up to that level.

c?ul: faster walk_1() for a set at a time

coul: bugfix - minp, maxp, midp should be 64-bit

This dates back to the first commit, when "-p" was only envisaged
for use as a first pass to restrict the upper limit.

coul: simplify level handling for walk_midp()

If we define the "normal" case as that in which the level is already
correct, then it turns out that the only normal case is on recovery.
The abnormal case is when we are triggered from apply_batch(), when
(naturally enough) the current level is that for the batch we just
applied.

coul: fix bug in recovery without -W

On recovery of a run without -W, we could wrongly think that it was
a run _with_ -W due to a missing guard.

sq12: formalize logging

As with pcoul, we write a start line, a progress line (every 600s)
and a finish line.

We do not attempt recovery at this point: an interrupted run should
be manually restarted after inspecting progress made.

coul: fix batch number in window title

When running a selected batch, we increment the batch number before
processing it. So allow for this when updating window title.

coul: use iterator instead of next_prime()

This makes 'pcoul -f7 -g3 12 9' about 10% faster.

coul: don't assume EINVAL for unsupported clock