[llvm] [AArch64] Pass scratch regs as operands to StoreSwiftAsyncContext. (PR #73332)

[Florian Hahn via llvm-commits]

fhahn wrote:

Discussed this also with @TNorthover offline. We could either just check if x16/x17 are not clobbered in `bool AArch64FrameLowering::canUseAsPrologue` and thus prevent ShrinkWrapping if they are ( instead of checking if there are any scratch registers)

Or default to X16/x17 if available, and if not pick alternatives

https://github.com/llvm/llvm-project/pull/73332