Design the Verona type system in MLIR

[rengolin]

Verona types are complex. Non-trivial pointer provenance, region encoding, mutability, data isolation, conditional execution all complicate the system. The language shouldn't have any more notation that is required to accurately infer all types at compile time, but that means a more complex type inference engine is needed.

Before we design that engine, we need to know how those types will be represented in MLIR, on both opaque and explicit representations.

Acceptance criteria:

• A (likely incomplete) consensus on the properties of the type system
• A quick example on how we convert from opaque to explicit and how we infer types across calls
• Examples in the opaque MLIR from Implement basic Parser #103 with tests updated to reflect design

Not included:

• Support for the explicit representation it in the parser
• Conversions or validation of any kind

[rengolin]

Initial implementation in #206