Support holdout data splits to be used in validation

[micahjsmith]

Can use the validation.split setting to specify a data split to be used in validation (i.e. test or holdout). The only significance of this setting is that data will be loaded using load_data(split=split) during validation. (Note that the pipeline and encoder will still be fit on the training data.) If your load_data function does not support loading a split by name, then ballet will fall back to using the training split.

[codecov-commenter]

Codecov Report

Merging #75 (541bae1) into master (5e419d4) will decrease coverage by 4.71%.
The diff coverage is 85.00%.

@@            Coverage Diff             @@
##           master      #75      +/-   ##
==========================================
- Coverage   84.38%   79.66%   -4.72%     
==========================================
  Files          47       54       +7     
  Lines        2753     2946     +193     
  Branches      281      282       +1     
==========================================
+ Hits         2323     2347      +24     
- Misses        341      506     +165     
- Partials       89       93       +4     

Impacted Files	Coverage Δ
ballet/client.py	45.83% <ø> (ø)
ballet/eng/misc.py	80.51% <ø> (ø)
ballet/validation/feature_acceptance/__init__.py	33.33% <0.00%> (ø)
ballet/validation/main.py	83.96% <84.00%> (-1.44%)	⬇️
ballet/__init__.py	100.00% <100.00%> (ø)
ballet/validation/base.py	90.69% <100.00%> (+0.22%)	⬆️
ballet/validation/feature_acceptance/validator.py	96.55% <100.00%> (ø)
ballet/validation/feature_pruning/validator.py	97.67% <100.00%> (ø)
ballet/validation/gfssf.py	93.84% <100.00%> (ø)
ballet/pipeline.py	83.33% <0.00%> (-12.50%)	⬇️
... and 8 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f72a053...541bae1. Read the comment docs.