BENCH add benchmark script for n_jobs=1

[tomMoral]

Add a benchmark script to test the overhead of calling joblib with n_jobs=1 vs calling a sequential loop for numpy matrix multiplications with various matrix sizes and for a very fast function (in red).

This gives

here, we can see that for small tasks, we have an overhead of roughly x10 per task for tasks that takes 1ms.
This state should be corrected by the changes pushed in #588.

[codecov]

Codecov Report

Base: 94.09% // Head: 94.04% // Decreases project coverage by -0.06% ⚠️

Coverage data is based on head (5dba7c2) compared to base (96c32ef).
Patch has no changes to coverable lines.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1400      +/-   ##
==========================================
- Coverage   94.09%   94.04%   -0.06%     
==========================================
  Files          52       52              
  Lines        7319     7319              
==========================================
- Hits         6887     6883       -4     
- Misses        432      436       +4     

Impacted Files	Coverage Δ
joblib/test/test_store_backends.py	94.54% <0.00%> (-3.64%)	⬇️
joblib/pool.py	87.80% <0.00%> (-0.82%)	⬇️
joblib/parallel.py	96.03% <0.00%> (-0.53%)	⬇️
joblib/memory.py	95.52% <0.00%> (+0.26%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[ogrisel]

Lgtm