Drop requirement for training data to have __len__

[mbatchkarov]

I'd like to be able to pass in the training data as a generator to reduce memory usage, which crashes sklear-crfsuite. Easiest fix is to remove len(X), but I also considered checking if X has a __len__ so people who load their data eagerly get a useful progress indicator

[codecov-commenter]

Codecov Report

Merging #52 into master will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master      #52   +/-   ##
=======================================
  Coverage   92.60%   92.60%           
=======================================
  Files           8        8           
  Lines         257      257           
  Branches       37       37           
=======================================
  Hits          238      238           
  Misses         15       15           
  Partials        4        4           

Impacted Files	Coverage Δ
sklearn_crfsuite/estimator.py	95.45% <100.00%> (ø)

---

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 7ada919...9bb92b6. Read the comment docs.