PCC: add basic "memory types".

[cfallin]

This PR adds a notion of "memory types" to the PCC (proof-carrying code) framework. To borrow a doc-comment:

A memory type is a struct-like definition -- fields with offsets, each field having a type and possibly an attached fact -- that we can use in proof-carrying code to validate accesses to structs and propagate facts onto the loaded values as well.

Memory types are meant to be rich enough to describe the layout of values in memory, but do not necessarily need to embody higher-level features such as subtyping directly. Rather, they should encode an implementation of a type or object system.

Note also that it is a non-goal for now for this type system to be "complete" or fully orthogonal: we have some restrictions now (e.g., struct fields are only primitives) because this is all we need for existing PCC applications, and it keeps the implementation simpler.

While developing this, we originally sketched out a more complete type-system that had arrays, allowed struct fields to be memory types themselves, and was in general more flexible. However, the possibility of cycles, non-statically-sized types, etc., was getting overly complex and I decided to YAGNI-simplify a bit for now. The Doc-comment on the module does sketch out what other types will be needed, and I think those plus a GV-bounded max fact will be enough to express PCC facts for all static and dynamic memories and table accesses in Wasmtime. (A partial further development of the more complete system, with some more validator checks for cycles, etc., is on this WIP branch.)

Part of this change was

Co-authored-by: Nick Fitzgerald fitzgen@gmail.com

[fitzgen]

r=me with nitpick-y suggestions that you can take or leave but also a more important point about a git submodule that should definitely be addressed before landing