Extending the indexing of GEM nodes

[achanbour]

This is the second part of PR #245 and concerns changes to GEM only. In summary, this adds new ways in which gem Nodes can be indexed, mimicking numpy's fancy indexing:

• indexing by an integer index array e.g., tensor_1d[1,2,3] using a new GEM index type called ListIndex` that wraps the index array and a free index that ranges over the elements of that array.
• supporting mixed indexing using arrays, slices, and indices e.g., tensor_2d[1::3, [2, 3, 4]].

[connorjward]

This appears to include a bunch of work from #246. I'm not sure what to review and what to ignore.

[achanbour]

This appears to include a bunch of work from #246. I'm not sure what to review and what to ignore.

Basically all the GEM stuff. I think the reason why it's showing the FIAT here is because I branched it off #246 which hasn't been merged yet.

Ultimately, we would want this PR to get merged after #246 since it would allow everything there to work on GEM nodes.

[achanbour]

Some remaining things to work on here include:

• Moving the indexing logic from getitem to gem.view()
• Unifying slicing and indexing with ListIndex in gem.view() -> produce a FlexiblyIndexed node?
• Possibly rewrite/simplify the construction of FlexiblyIndexed