feat: support runtime model index N in expressions and indexing

[martinjrobins]

Summary

• add reserved model index as a runtime scalar
• extend integer-index expressions to support and allow only as a name
• support in general expressions in LLVM and Cranelift codegen
• add across compiler interfaces/call paths and preserve ordering of UInt args
• add parser/discretiser/compiler tests, including failure for defining tensor

see #112

[martinjrobins]

need to change the API of the model_index param. Remove this from rhs, stop etc, and add it to SetInputsFunc. The model_index pass in here will be written to a global value in the codegen module, ready to be used during the other functions. So the API is (a) user sets the model index using set_inputs once, then (b) this is used in subsequent u0, rhs, stop etc functions.

So N is no longer time dependent, but is not input dependent