Fix bug within _run_thread_job

[AlessandroMiola]

Hi @cgnorthcutt,
this PR is intended to fix the incorrect behaviour highlighted within this issue wrt the handling of the best estimator and whose solution was somehow suggested in the issue itself. Indeed, it can be noticed that also sklearn cross-validation functions (see here) do clone the estimators before fitting (thus training a clone on each fold so as to make sure all folds are independent).

Eventually, it still does not solve the problem arising whenever parallelize parameter is set to True (default).

[codecov]

Codecov Report

Merging #26 (0b5daaf) into master (0bea8a0) will decrease coverage by 1.08%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #26      +/-   ##
==========================================
- Coverage   72.78%   71.69%   -1.09%     
==========================================
  Files           3        3              
  Lines         158      159       +1     
==========================================
- Hits          115      114       -1     
- Misses         43       45       +2     

Impacted Files	Coverage Δ
hypopt/model_selection.py	71.15% <100.00%> (-1.11%)	⬇️

---

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 0bea8a0...0b5daaf. Read the comment docs.

[cgnorthcutt]

Looks good. Thank you for your contribution @AlessandroMiola and for mentioning that the parallelization issue still remains. Merging.