When calibrating with multiple timeblocks, the start, end and average
times of the timeblocks are written to the hyperdrive-formatted
calibration solutions. However, they weren't written correctly. Using
shorter key names for these times (e.g. "START_TIMESTAMPS" -> "S_TIMES")
appears to fix the issue. I don't fully understand why this works; maybe
there's probably a bad interaction in cfitsio between long key names and
long strings.

In addition, the way that solutions were propagated across timeblocks
was incorrect; this is now fixed.

Multiply visibilities by weights before calibration and divide by
weights if writing visibilities out. This means weights are needed
during calibration and the code is able to run about 25% faster.

Note that there are relatively big (up to 1e-4) float-point errors
introduced by doing this; writing visibilities out of di-calibration
should therefore only be done as a "quick and dirty" convenience. When
solutions-apply is introduced, that will be the preferred way to obtain
calibrated visibilities.

Autocorrelations will only be calibrated and written out as part of a
to-be-written solutions-apply command.

Fix one test passing --cpu to SimulateVisArgs when it was not expected.

The parse_{time,freq}_average_factor functions expect {time,freq}
resolutions, but they weren't getting them, and things weren't working
as expected. This commit fixes that and an associated test.

round jd_frac before it gets added to PZERO5
add round_hundredths_of_a_second_duration

-> implement time averaging on calibrate-vis output when the
selected timesteps are non-contiguous.

di-calibrate cannot currently write its sky model to an ms -- the help
text has been updated to account for this and a test now exists to check
that the help text is correct.

There were problems around files and directories being overwritten. I've
done the best I can but it looks like we can't ever handle detecting
whether we have permission to remove a directory ahead of time. We may
want to improve an error message associated with attempting to remove a
directory we don't have permission to, if it's not already clear enough.

A message stating where calibrated visibilities have been written to was
lost; this is now back.