CAS-13901 : Fix asp tests in deconvolve tests
Edited the passing thresholds for the asp tests in test_task_deconvolve.
Used a smalled gain value (gain=0.2) similar to test_task_tclean, and
the passing criterion are based on the values for a single major cycle
in test_task_tclean.
Also enabled asp testing on Darwin - with a lower gain it is expected to
be more stable across different systems.
Pull request #767: CAS-13791
Merge in CASA/casa6 from CAS-13791 to master
* commit '0b3b01d988822e881a47dbb8a9138d5622ebfae6': CAS-13791 : Modified tests to have gain=0.2. This reduces variability across build systems to below the standard tolerance threshold. Enabled the test for Mac too. CAS-13791 : edited the ver test func for asp CAS-13791: Set 'fusedthreshold = -1' to allow AspClean never switches to Hogbom. This makes G55 result even better. CAS-13791: cherry-pick from ARD-31 for adding "scaling factor" to bfgs to improve bfgs optimizations when parameters are with widely different magnitudes. With this fix, G55 now runs fine and it does not affect behaviors on other datasets.
CAS-940-improve: Improve computational efficiency by 1) measuring rms from masked residual and 2) move the initialization of blc and trc of Dirty to the beginning since it only needs to be done once. This improves runtime by 6mins to 3mins (with mask) and 3mins to 2:30 (no mask). Note that the runtime here is just one run and the runtime may vary between runs.
CAS-940-improve: fixed a bug in creating initial scales.
