Cranelift: Support loading and storing vector types smaller than 128bit

#### Feature

While Cranelift can represent vector types smaller than 128bit, load and store operations for these smaller vector types are not implemented.

#### Benefit

Rustc has started passing vector types in vector registers on x86_64. Some of these vector types may be smaller than 128bit. Previously these vector types didn't need to be exposed to Cranelift as all vector operations are emulated by cg_clif using scalar operations. However now that they are passed in vector registers, cg_clif needs to use vector types at the ABI level and thus needs to support loading and storing small vectors.

#### Implementation

TODO: Do you have an implementation plan, and/or ideas for data structures or
algorithms to use?

#### Alternatives

Have a bunch of code in cg_clif dedicated to handling this. I probably need this anyway for the time being, but I prefer if I can get rid of it as soon as possible through native Cranelift support.