Finish implementing support for array shapes in type checker

[TysonAndre]

related to #1343

class ArrayShapeType exists for array{field:T}, but I still need to do the following:

• Stop immediately casting it to a GenericArrayType (T[])

• When parsing union types from AST nodes, generate ArrayShapeType instead of GenericArrayType based on a heuristic. (heuristic TBD)

  E.g. if one of the key values can't be determined, then convert to GenericArrayType.

• Add rule to check if an ArrayShapeType can cast to another ArrayShapeType, add unit tests

• Add a rule to check if a GenericArrayType can cast to an ArrayShapeType, and vice versa

• When normalizing union types, make sure there is at most one ArrayShapeType.
  (Haven't created a specification of what to do when there are both GenericArrayType and ArrayShapeType. One possibility is to use the GenericArrayType when keys don't match)

• Merge two ArrayShapeType into a single ArrayShapeType when there are two branches

• Figure out and implement a heuristic for normalizing types when recursively analyzing function calls for when quick_mode is off (one heuristic is to always cast to GenericArrayType when recursing to analyze).

• In an assignment, If the key value is known, AssignmentVisitor should update GenericArrayType with the new key value

• (Optional) support unset (Doesn't catch that many bugs)

• ArrayReturnType plugin will need to be changed to properly handle array shapes.

• UnionTypeVisitor will need to check the values of keys during array accesses. If a GenericArrayType is part of the union type, and contains that key, then the corresponding field type from the shape should be returned.

---

Follow-up tasks:

• Indicate fields are optional in a shape. Modify the type checker and merging logic. Warn about fetching an optional field.

Other followup tasks:

• Add more unit tests
• Improve scalar casting rules. E.g. array{a:int} should not be able to cast to array{a:int, other:int}.
  A followup PR may add optional fields to array shapes
• Test this on large projects for crashes and unexpected results
• Reduce false positives in loops (Another PR provided ways to suppress issues for an entire file (phan-file-suppress) or to set a variable's type inline (phan-var in top level string literal))
• Look into inferring [] as array{} in future PRs. This was disabled due to high false positives in loops