MRG: Clean up and unify _prepare_forward

[larsoner]

This PR should not change any outputs, but internally it:

• Refactors _prepare_forward and _check_loose_forward to greatly simplify the logic they use, and deduplicate checks/flows across MNE and sparse solvers.
• Adds ordered argument to pick_channels and pick_channels_forward, which is something that I have often wanted. Eventually we can probably simplify a lot of our code using ordered=True, as there are a lot of places we do the same picking-with-order logic instead of picking-as-set logic.
• Fixes the CircleCI regression in master.
• Deprecates the old way of using compute_depth_prior in favor of a cleaner, Forward-based method.
• Adds an ignore for scanno in the inv round-trip test (should not matter, and I think our result is now possibly more correct because it resets scanno to match the number of entries in info)

These are important for #5881, and are hopefully the last steps we need (along with #5979) before having easy, unified, drop-in lead-field normalization across all solvers (!).

Ready for review/merge from my end @agramfort

[codecov]

Codecov Report

Merging #5999 into master will increase coverage by 0.01%.
The diff coverage is 86.46%.

@@            Coverage Diff             @@
##           master    #5999      +/-   ##
==========================================
+ Coverage   88.91%   88.93%   +0.01%     
==========================================
  Files         401      401              
  Lines       72816    72833      +17     
  Branches    12108    12108              
==========================================
+ Hits        64746    64773      +27     
+ Misses       5184     5179       -5     
+ Partials     2886     2881       -5

[larsoner]

This pull request introduces 1 alert when merging 4ee9f7c into f7bb2ef - view on LGTM.com

new alerts:

• 1 for Redundant comparison

---

Comment posted by LGTM.com

[larsoner]

Merged after chat with @agramfort

[agramfort]

Thx @larsoner