Fix Learner._check_input_formatting() for dense feature sets.

[desilinguist]

This PR closes #656.

• Right now, this method basically assumes that all featuresets will be sparse and, hence, uses the .data attribute to get a flattened version of the feature array. However, this does not work for featuresets with dense feature arrays (see the issue for a detailed description of the problem).
• To handle dense feature arrays, we can use the .flat attribute that returns a 1-dimensional iterator over the dense array.
• Added three new tests:
  • 2 tests that train SKLL models -- one regression & one classification -- with a non-sparse featureset (which would not have worked before this PR)
  • 1 to trigger the other check from this method -- regression with string labels -- which is not something we are testing for currently.

[codecov]

Codecov Report

Merging #658 (a9321a4) into main (32823c0) will increase coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #658      +/-   ##
==========================================
+ Coverage   95.06%   95.09%   +0.03%     
==========================================
  Files          27       27              
  Lines        3098     3101       +3     
==========================================
+ Hits         2945     2949       +4     
+ Misses        153      152       -1     

Impacted Files	Coverage Δ
skll/learner/__init__.py	96.47% <100.00%> (+0.19%)	⬆️

---

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 32823c0...a9321a4. Read the comment docs.