AMDGPU: Always custom lower extract_subvector

The patterns were ripped out in
a4a3ac1 so this always needs to be
custom lowered. I absolutely hate how difficult it is to write tests
for these, I have no doubt there are more of these hidden.

Fixes #64142

(cherry picked from commit 95e5a46)

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

@@ -278,10 +278,10 @@ SITargetLowering::SITargetLowering(const TargetMachine &TM,
       case ISD::UNDEF:
       case ISD::EXTRACT_VECTOR_ELT:
       case ISD::INSERT_VECTOR_ELT:
-      case ISD::EXTRACT_SUBVECTOR:
       case ISD::SCALAR_TO_VECTOR:
       case ISD::IS_FPCLASS:
         break;
+      case ISD::EXTRACT_SUBVECTOR:
       case ISD::INSERT_SUBVECTOR:
       case ISD::CONCAT_VECTORS:
         setOperationAction(Op, VT, Custom);