Compare like pols in test_gain

[james-smith-za]

The qualification/tied_array_channelised_voltage/test_gain test occasionally shows a difference one greater expected for a single value in the test every now and then. It's very intermittent. To a first approximation this shouldn't be possible because the tolerance for each test should be taken into account and there's no way for dithering to cause a bigger difference assuming that the incoming signal is the same.

We think that this is caused by separate dithering for each pol in the F-engine which means that despite having the same d-sim input, the F-engine output for pol0 and pol1 may be ever so slightly different from each other, just different to cause the error.

In this commit I adjust the test to using the same pol in three different beams for comparison, rather than three half-beams which would have been pol0, pol1, pol0.

This appears to solve the problem.

Closes NGC-1473.

Checklist (if not applicable, edit to add (N/A) and mark as done):

• (N/A) If dependencies are added/removed: update setup.cfg and .pre-commit-config.yaml
• (N/A) If modules are added/removed: use sphinx-apidoc -efo doc/ src/ to update files in doc/
• Ensure copyright notices are present and up-to-date
• If qualification tests are changed: attach a sample qualification report No significant change in report.
• (N/A) If design has changed: ensure documentation is up to date
• (N/A) If ICD-defined sensors have been added: update fake_servers.py in katsdpcontroller to match

[bmerry]

It looks like you're capturing 6 half-beams, but then ignoring 3 of them. I'd suggest either
(a) set stream_names to grab only the 3 half-beams you're interested in; or
(b) update the comparison code to test the odd-indexed polarisations against each other.

I'd lean towards option (b) because it's a stronger test (ensures we haven't mucked up gain-setting such that it only applies to one polarisation).