divergence-resolution: use last evolution date to choose p1 when merging

Before this patch, we choose the minimum revision as p1 while merging the
two divergent csets which had a drawback that if independent user resolve
the same divergence, their final resolved cset would have different hashes
(because of 'divergence_source_local' and 'divergence_source_other' extras).

Now, we decide the p1 on the basis of which of the two divergent cset
was rewritten more recently. This new logic removes the "different hash"
problem.

To save us from big output changes in the tests due to this change, I also
added the second factor i.e revision_number while sorting the csets wrt dates
(as date is same for all the csets in tests) to fallback to the old way of
picking the revision i.e choosing the minimum rev number.

And to demonstrate that now divergence resolution is independent of which
side user run the `hg evolve --content-div` resolved cset id won't change,
I have added a separate test file.
Flag --config devel.default-date='...' is being used here to record a
custom date in the obsmarkers.
8 jobs for topic/default/resolution-order in 4 minutes and 33 seconds (queued for 2 seconds)
Status Job ID Name Coverage
  Test
passed #190612
checks-py2

00:02:02

passed #190613
checks-py3

00:01:19

passed #190618
doc

00:00:36

manual #190611
allowed to fail manual
pytype
passed #190614
tests-py2-cext

00:04:32

passed #190615
tests-py2-pure

00:02:46

passed #190616
tests-py3-cext

00:04:30

passed #190617
tests-py3-pure

00:04:05