MRG, ENH: Add Label.restrict and faster patch info

[larsoner]

1. Add label.restrict, the opposite of label.fill
2. Add dist_limit=0. to allow using faster dijkstra calculation in SciPy 1.3+ to only calculate patch information
3. Remove errant statement in docs of n_jobs parameter of add_dist and dist_limit about parallelization depends on the number of source spaces (it parallelizes over vertno actually).
4. EDIT: Add test infrastructure workarounds for nilearn using deprecated sklearn stuff hitting us due to sklearn 0.22 release

[codecov]

Codecov Report

Merging #7108 into master will increase coverage by 0.01%.
The diff coverage is 93.75%.

@@            Coverage Diff             @@
##           master    #7108      +/-   ##
==========================================
+ Coverage    89.7%   89.71%   +0.01%     
==========================================
  Files         444      444              
  Lines       78976    79019      +43     
  Branches    12674    12677       +3     
==========================================
+ Hits        70848    70895      +47     
+ Misses       5317     5315       -2     
+ Partials     2811     2809       -2

[larsoner]

FYI this now contains some testing fixes that fix failures in current master (see, e.g., #7109) due to sklearn v0.22 coming out, and related nilearn deprecated usage (fixed upstream but not released yet)

[larsoner]

In that case I'll see if I can fix the docstring checking code, then merge once CIs are happy

[agramfort]

sounds good thx

…

[larsoner]

There are a lot of errors to fix :( I'll fix the one introduced here, merge, then fix the others in a separate PR.