Add initial cursor skeleton

[alexreinking]

This PR lays the foundation for building a new scheduling API on top of our cursors idea. In particular, this PR includes:

1. A new abstract Cursor type that has three concrete subtypes: a Node, a Block, and a Gap.
2. Public navigation methods on these objects to help select parents, children, and gaps surrounding nodes and blocks.
3. Internal functions for making unsafe AST changes that produce a forwarding function as collateral.
4. Modifications to the pattern matching system that return cursors instead of (lists of) raw LoopIR
5. A bugfix in the pattern matching system. Matching a single statement now always returns a block containing that single statement, rather than the statement and the remainder of the block.

Although cursors are exposed through the public API, they are currently not very useful. Only the navigation functions are public, as is the underlying proc getter. Otherwise, no other public APIs accept them.

This type will ultimately serve (at least) three purposes:

1. It is a forwardable handle for scheduling directives, such that our vision of user-defined scheduling directives may be realized.
2. It is an introspection tool for users to make particular scheduling decisions as in (1)
3. It may be used internally for efficient AST modification and simplified forwarding definitions.

No progress is made on (2) in this PR.