RF: Avoid import-time loading of nose, mock and nibabel.testing

[effigies]

Related to discussion in #698.

h5py and pydicom would be more complicated, due to the way optpkg and pydicom_compat are implemented, and nose and mock constituted over half of the load time (with numpy preloaded before calculating).

[coveralls]

Coverage increased (+0.02%) to 91.839% when pulling 937cd9d on effigies:rf/delayed_imports into 4b6ca81 on nipy:master.

[effigies]

Should be ready to go, if anybody wants to have a review. It's pretty minimal, the biggest trouble being working around some mock magic.

[codecov-io]

Codecov Report

Merging #699 into master will not change coverage.
The diff coverage is 78.94%.

@@          Coverage Diff           @@
##           master    #699   +/-   ##
======================================
  Coverage    89.1%   89.1%           
======================================
  Files          93      93           
  Lines       11469   11469           
  Branches     1990    1992    +2     
======================================
  Hits        10220   10220           
+ Misses        910     908    -2     
- Partials      339     341    +2

Impacted Files	Coverage Δ
nibabel/testing/__init__.py	94.79% <ø> (+0.79%)	⬆️
nibabel/nicom/dwiparams.py	71.69% <100%> (ø)	⬆️
nibabel/casting.py	84.74% <100%> (ø)	⬆️
nibabel/quaternions.py	99.02% <100%> (ø)	⬆️
nibabel/affines.py	100% <100%> (ø)	⬆️
nibabel/nifti1.py	91.22% <100%> (ø)	⬆️
nibabel/optpkg.py	90% <60%> (ø)	⬆️
nibabel/__init__.py	91.48% <77.77%> (-1.54%)	⬇️

---

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 4b6ca81...937cd9d. Read the comment docs.