Support voliatile memory intrinsics

[Yawning]

Is your feature request related to a problem? Please describe.

I need a way to zero memory that will not be removed by the various forms of optimization.

Describe the solution you'd like

The llvm.memcpy/llvm.memmove/llvm.memset intrinsics calls have a isvolatile parameter that is currently not exposed. To the best of my knowledge, implementing a call analogous to memset_s (C11 Annex K)/SecureZeroMemory (Windows) can be done on LLVM by using a volatile llvm.memset followed by a sequentially consistent memory fence (atomic_fence() from core:intrinsics).

For my purposes being able to do a volatile llvm.memset is sufficient, but others may want volatile versions of the other functions.

Additional context

Dead Store Elimination (Still) Considered Harmful gives an overview of the problem space and common techniques (some which are flawed) to address this problem.

The zeroize Rust crate uses the volatile memeset + barrier approach.