Operational semantics for allocator elision

https://github.com/llvm/llvm-project/pull/177592 specifies allocation elision in terms of allowed transformation.

The resulting semantics are angelic in nature, as they require knowing whether a free with the same alloc-family occurs in the future. Ideally, we would find a way to make these semantics operational.

The thing that makes this non-trivial is that the free call may be inlined, in which case allocation elision can no longer be applied.

Start by reading PR 177592, which specifies allocation elision as an allowed transformation. Examine how its semantics depend on a future free with the same alloc-family and how inlining affects whether elision can be applied. Done means defining an operational alternative that addresses the inlined-free case.

Written by the indexing model from the issue text.