Investigate using memory use markers in the generated LLVM

LLVM has intrinsics like `llvm.lifetime.start` and `llvm.lifetime.end` which are used for explicitly specifying a stack object's lifetime. If we use it, the LLVM `opt` tool might figure out more of our invariants regarding stack-allocated objects and might be able to optimise them better. (For example, decide to let two non-intersecting variables share the same chunk of stack space.)

Here's the reference: https://llvm.org/docs/LangRef.html#memory-use-markers

The changes would be applied at every `alloca`, so at the very least in the `case machine.Allocate` of the `transform` function (as far as I can tell, the lifetime is after extracting the local reference & before the terminator), and in the push-/popEnvironment functions. We should also be able to use `llvm.invariant` intrinsics to specify that a piece of memory won't change.