Investigate and fix test fails bypassed in PR 347

[mcgarth]

PR #347 introduced a crucial bug fix to correct the way unit conversion is applied. However, these changes introduced test fails in the test_mpi_vs_multiprocess_vs_single_process module.

The tests were failing in the tests.common.assert_tifs_equal function. Some were failing when CVD_MAXVAR and/or CVD_ALPHA were not equivalent to 1.d.p. between mpi/multiprocess/serial jobs:

PyRate/tests/common.py

Line 196 in 8893509

assert round(eval(md_sds[k]), 1) == round(eval(md_mds[k]), 1)

. Others were failing where interferogram arrays were not equivalent to 3.d.p. as in the current test:

PyRate/tests/common.py

Line 203 in 8893509

np.testing.assert_array_almost_equal(d1, d2, decimal=3)

or even 1.d.p. when the threshold was relaxed.

To bring the unit conversion bug fix in to master ASAP, the test fail problem was bypassed in commit 0a5363e by switching these tests to use the old unit test data (gamma_conf) rather than the new Mexico "CropA" data (mexico_cropa_conf).

The reasons for the test fails using Mexico CropA data should be fully investigated and rectified.