Cranelift: Support stack probes without external function call

[bjorn3]

Feature

LLVM has support for performing the stack probes directly in the function prologue instead of calling an external function. (see rust-lang/rust#77885 for using this in Rust)

Benefit

This can be faster due to being able to unroll the stack probe loop. It also avoids having to provide an external stack probe function. This makes it possible to enable stack probing in cg_clif. (https://github.com/bjorn3/rustc_codegen_cranelift/issues/661)

Implementation

This should only be implemented for the machinst backends. It should probably be added to either the gen_prologue implementation or gen_prologue caller.

Alternatives

Keep using an external stack probe function with the associated drawbacks.

[bjorn3]

Neither AArch64, nor s390x has support for this yet.

[afonso360]

Right, should we reopen this or open issues against both backends?

[bjorn3]

Opened #4846 and #4847.