P Jones rotating the wrong way around

[bennahugo]

See https://github.com/ska-sa/meqtrees-cattery/blob/master/Cattery/Siamese/OMS/feed_angle.py#L140

As per @bmerry's comment this is the wrong way around @sjperkins please note

After doing some experiments I'm tentatively asserting that Meqtrees
is doing it wrong. The experiment involved simulating a single point
source at phase centre, with I=3Jy Q=1Jy U=V=0, at dec -85° (close to
the SCP so that parallactic angle is mostly dominated by sky rotation
while az/el are relatively constant), as seen from MeerKAT, and using
just the PJones term. I then plotted the observed XX, YY and XY
visibilities and the parallactic angle (as provided by
addDerivedMSCal) over time. I then
- worked out how these should rise and fall relative to each other
based on some diagrams of the sky rotating
- worked it out again based on the equations in Ben's calibration report
- computed what the visibilities should be based on the formula for P
in Hales 2017 (which I also verified).
and all of the above seem to indicate that Meqtrees is rotating in the
wrong direction.

Looking at the source, I notice the following in turbo-sim.py, which
suggests that the implementation was changed at some point (git blame
describes the commit as "added P that is consistent with
calico-stefcal":

meqmaker.add_uv_jones('P','feed orientation',[feed_angle]);
# meqmaker.add_uv_jones('P','feed angle',Rotation('P'));

and feed_angle seems to implement the inverse operation to Rotation:
it uses "Meq.Matrix22(cos,-sin,sin,cos);" whereas Rotation uses
"Meq.Matrix22(cos,sin,-sin,cos);".

[bennahugo]

I can confirm this is a bug - the other way around is implemented in cubical
https://github.com/ratt-ru/CubiCal/blob/pol-cal/cubical/machines/parallactic_machine.py#L135
and gives consistent results with CASA

[bennahugo]

@o-smirnov lets get a sign fix for this in before making a release that officially supports polcal