Add a `picks` parameter to `vector_auto_regression` and `indices` to `envelope_correlation`

[tsbinns]

Describe the problem

vector_auto_regression and envelope_correlation will compute results for all channels, meaning users must first pick channels before passing this to the functions. This goes against the API of similar functions in MNE where a picks parameter is present (or indices in MNE-Connectivity).

Also since #334, bad channels will always be excluded from computations in these functions, but many other functions in MNE allow the picks (or indices) parameter to have users explicitly include bads.

Describe your solution

Add a picks parameter to vector_auto_regression and indices to envelope_correlation, with the same functionality as elsewhere in MNE.

I'd say picks is more appropriate for vector_auto_regression than indices, since all-to-all connectivity is implied for VAR models.