PR nilearn#1271 introduces this regression where ensure_finite=True modifies the data inplace instead of copying it (nan_to_num does an implicit copy). I preferred to get back to nan_to_num instead of just doing a copy because I believe that relying on numpy functions is safer and more future proof.

…e=True

…clean

* 'master' of https://github.com/nilearn/nilearn: (2100 commits)
  Fixes nilearn#2029 Handle gzip files without extensions (nilearn#2126)
  Update Schaefer parcelation to v0.14.3 (nilearn#2138)
  MAINT: Future-compatible cmap reversal (nilearn#2131)
  fix openmp crash (nilearn#2140)
  change nose to pytest on appveyor (nilearn#2130)
  adding fix to whatsnew
  fix wrong urls in nki dataset
  Made Flake8 happy
  added whats_new
  Added test for labels=None
  More explicit doc string
  Test for right length of labels
  Added tests for empty list case and labels as ndarray
  ensure labels is a list, not an ndarray
  Change such that empty lists are recognized as False
  Undid some documentation changes as their reason is not understood
  Replaced conda latest version with 4.6.* pending replacement
  Removed omit sections (nothing to omit, it would seem)
  Corrected docstring for bg_img in plot_stat_map
  Removed omit=*test* to fix error during coverage reporting
  ...

# Conflicts:
#	nilearn/tests/test_signal.py