Fix numpy related type issues

[jklaise]

This PR is a prerequisite for #521 as the newest tensorflow pulls in the newest numpy which has support for type hints. This has revealed some typing issues in our codebase.

This PR uses type: ignore fairly liberally as in some places it is not fully clear why the upstream types are failing (e.g. List[int] not being valid input to np.ix_), and in others we're missing some validation/type narrowing on our side (e.g. baselines). I've also used cast in a few places to help the typechecker narrow down types based on the boolean flags we use.

[codecov]

Codecov Report

Merging #543 (3588fc1) into master (dbc874a) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #543   +/-   ##
=======================================
  Coverage   82.47%   82.47%           
=======================================
  Files          77       77           
  Lines       10417    10420    +3     
=======================================
+ Hits         8591     8594    +3     
  Misses       1826     1826           

Impacted Files	Coverage Δ
alibi/explainers/anchor_image.py	91.30% <ø> (ø)
alibi/explainers/anchor_tabular.py	78.75% <100.00%> (ø)
alibi/explainers/integrated_gradients.py	89.00% <100.00%> (+0.08%)	⬆️
alibi/explainers/tests/test_anchor_image.py	97.93% <100.00%> (ø)