Filter elements with zero sample size in fathpd

[sfiligoi]

Elements with zero sample size have no clear semantics in faithpd, so we filter them out.
But there was a historical assumption that a single zero element is allowed and would evaluate to 0, so we preserve that option, but issue a warning.

[sfiligoi]

This addresses #51

[sfiligoi]

CC @vpa1977

[sfiligoi]

I checked that the unifrac python tests still work.

@vpa1977 Can you check if this fixes the problems in #51 for you, since I cannot reproduce it myself?

[vpa1977]

Thank you!!!!

I am getting

$python3 unifrac/tests/test_api.py 

WARNING: All samples had zero counts. Forcing zero result.
.
----------------------------------------------------------------------
Ran 1 test in 0.014s

OK

and in the debugger everything looks right.

I wonder if it would make sense to add some test for "WARNING: Some samples had zero counts and were filtered out.\n", or it is an edge condition that does not occur in the reality?

[wasade]

Unlikely in real use, thanks!