Fix bda overload to return an implementation

[sjperkins]

• Tests added / passed

  $ py.test -v -s africanus

  If the pre-commit tests fail, install and
  run the pre-commit hooks in your development
  virtuale environment:

  $ pip install pre-commit
  $ pre-commit install
  $ pre-commit run -a

• Fully documented, including HISTORY.rst for all changes
  and one of the docs/*-api.rst files for new API

  To build the docs locally:

  pip install -r requirements.readthedocs.txt
  cd docs
  READTHEDOCS=True make html

[landmanbester]

Is there a reason why chan_freq is not returned?

[sjperkins]

Good question. Part of the reason is that it technically transforms to (row, chan) dimensionality as the frequencies and channel widths can also vary per row. This sort of data is available on the meta object, higher up in this namedtuple construction. There's a chan_width member with (row, chan) dimensionality. However, there isn't a chan_freq member. Do you need it? I suppose it wouldn't be necessary if one were only doing time BDA.

[landmanbester]

Indeed, I don't need it for this reason but just wanted to make sure it's not an oversight.

[sjperkins]

Ah there's already an issue

• Support FREQ_CENTROID and EFFECTIVE_BW per sample in BDA code #263

[landmanbester]

I should add that when doing time only BDA you shoudn't need uvw to be expanded as rowchan. This actually uses quite a bit of memory. It's not a train smash for now but it might be worth adding a time only BDA function in the future that doesn't do this

[landmanbester]

LGTM. I've tested this on some simulations and at least the time only case seems to work as expected. Thanks @sjperkins with that we now have BDA support in pfb