Remove scope types

[samestep]

Previously we've had a special type called Scope to differentiate function-local Refs with different scopes, so that they can't be assigned to each other willy-nilly, and so that we can prevent usage of a reference when it goes out of scope. This PR removes those so now all Refs with the same inner type are treated the same.

The reason for this is that we need to treat scopes slightly more flexibly in #86; specifically, if you Select on two Refs, it's unclear which scope the result should have. It can't have its own new scope, because having its own scope means that it can be Resolved later. But that problem goes away if both Refs are just the same type. Also, this removes an additional pass in transposition where we would need to keep track of what variables need to have new scope types added for them: with no scope types, that sort of self-referential-ness goes away.

The reason we originally had scope types at all was because it seemed like differentiating them at the type level would make scopes easier to validate. But we already have to validate scopes of other kinds of variables, and this doesn't really seem different. All you need to do is ensure that Refs introduced in a block get resolved later in that same block. Then for derived Refs, just have them point to the most recently defined non-derived Ref that they use; this arises in Select, for instance. We already use this most-recently-defined idea in rose-web for normal variables to deal with lazily-defined literals. Also, note that this behavior for Select constrains us to have scope nesting be properly matched: for instance, it wouldn't work to define two Refs and then later resolve them in non-reverse order, because then a Selected Ref would only get invalidated after the second one rather than the first one.