Add observed option for speed with many bins

[benkrikler]

When there are many bins in a binned dataframe producing all of them, even those that aren't actually filled, can consume a lot of memory and a lot more processing time. This new option enables the resulting dataframe to be much more sparse by only filling bins with non-zero. This can be combined with the pad_missing option that existed previously to force these bins to be added back in the merging step.

Note that in the future this option might be automatically enabled / disabled in conjunction with pad_missing since this is really a performance optimisation control that a user often may not want to have to know about.

[codecov]

Codecov Report

Merging #118 into master will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #118      +/-   ##
==========================================
+ Coverage   65.38%   65.40%   +0.02%     
==========================================
  Files          20       20              
  Lines        1456     1457       +1     
==========================================
+ Hits          952      953       +1     
  Misses        504      504              

Impacted Files	Coverage Δ
fast_carpenter/summary/binned_dataframe.py	85.71% <100.00%> (+0.08%)	⬆️
fast_carpenter/version.py	100.00% <100.00%> (ø)