[llvm-mca] Improve barriers for strict region marking (PR52198)

As suggested on the bug, to help (but not completely....) stop folded instructions crossing the inline asm barriers used for llvm-mca analysis, we should recommend tagging with memory captures/attributes.

Differential Revision: https://reviews.llvm.org/D117788

Author: RKSimon

llvm/docs/CommandGuide/llvm-mca.rst

@@ -299,9 +299,9 @@ C++. As a workaround, inline assembly directives may be used:
 .. code-block:: c++
 
   int foo(int a, int b) {
-    __asm volatile("# LLVM-MCA-BEGIN foo");
+    __asm volatile("# LLVM-MCA-BEGIN foo":::"memory");
     a += 42;
-    __asm volatile("# LLVM-MCA-END");
+    __asm volatile("# LLVM-MCA-END":::"memory");
     a *= b;
     return a;
   }