Lock free max / min pattern for CPU

[NN---]

JIT should detect CAS loop pattern for max / min and replace it with CPU instruction id available.

int current;
do
{
    current = Volatile.Read(ref back);
    if (current >= value)
        return;
} while (Interlocked.CompareExchange(ref back, value, current) != current);

For example this or similar code pattern on ARM this should produce LDUMAX:

https://developer.arm.com/documentation/ddi0596/2021-06/Base-Instructions/LDUMAX--LDUMAXA--LDUMAXAL--LDUMAXL--Atomic-unsigned-maximum-on-word-or-doubleword-in-memory-

[gfoidl]

For reference: #121264

[huoyaoyuan]

It's much more complex to detect such pattern since multiple branches are involved.

[NN---]

Then intrinsic is the better approach.