[msan] Handle Arm NEON pairwise min/max instructions (#129824)

Change the handling of:
- llvm.aarch64.neon.fmaxp
- llvm.aarch64.neon.fminp
- llvm.aarch64.neon.fmaxnmp
- llvm.aarch64.neon.fminnmp
- llvm.aarch64.neon.smaxp
- llvm.aarch64.neon.sminp
- llvm.aarch64.neon.umaxp
- llvm.aarch64.neon.uminp
from the incorrect heuristic handler (maybeHandleSimpleNomemIntrinsic)
to handlePairwiseShadowOrIntrinsic.

Updates the tests from #129760

Adds a note that maybeHandleSimpleNomemIntrinsic may incorrectly match
horizontal/pairwise intrinsics.

Author: thurstond

llvm/lib/Transforms/Instrumentation/MemorySanitizer.cpp

@@ -3129,6 +3129,9 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
   /// fine).
   ///
   /// Caller guarantees that this intrinsic does not access memory.
+  ///
+  /// TODO: "horizontal"/"pairwise" intrinsics are often incorrectly matched by
+  ///       by this handler.
   [[maybe_unused]] bool
   maybeHandleSimpleNomemIntrinsic(IntrinsicInst &I,
                                   unsigned int trailingFlags) {
@@ -4871,6 +4874,18 @@ struct MemorySanitizerVisitor : public InstVisitor<MemorySanitizerVisitor> {
       setOrigin(&I, getCleanOrigin());
       break;
 
+    // TODO: handling max/min similarly to AND/OR may be more precise
+    // Floating-Point Maximum/Minimum Pairwise
+    case Intrinsic::aarch64_neon_fmaxp:
+    case Intrinsic::aarch64_neon_fminp:
+    // Floating-Point Maximum/Minimum Number Pairwise
+    case Intrinsic::aarch64_neon_fmaxnmp:
+    case Intrinsic::aarch64_neon_fminnmp:
+    // Signed/Unsigned Maximum/Minimum Pairwise
+    case Intrinsic::aarch64_neon_smaxp:
+    case Intrinsic::aarch64_neon_sminp:
+    case Intrinsic::aarch64_neon_umaxp:
+    case Intrinsic::aarch64_neon_uminp:
     // Add Pairwise
     case Intrinsic::aarch64_neon_addp:
     // Floating-point Add Pairwise