[llvm] [SPIR-V] Emit valid SPIR-V code for integer sizes other than 8,16,32,64 (PR #94219)

[Nathan Gauër via llvm-commits]

================
@@ -10,7 +10,10 @@
 ; CHECK-SPIRV: OpCapability BitInstructions
 ; CHECK-SPIRV: OpExtension "SPV_KHR_bit_instructions"
 ; CHECK-SPIRV: %[[#CharTy:]] = OpTypeInt 8 0
-; CHECK-SPIRV: %[[#]] = OpBitReverse %[[#CharTy]] %[[#]]
+; CHECK-SPIRV-NO: %[[#CharTy:]] = OpTypeInt 8 0
+; CHECK-SPIRV: %[[#Arg:]] = OpFunctionParameter %[[#CharTy]]
+; CHECK-SPIRV: %[[#Res:]] = OpBitReverse %[[#CharTy]] %[[#Arg]]
----------------
Keenuts wrote:

Isn't there a behavior change between the LLVM-IR to SPIR-V?
-> LLVM IR: input='0b10', output='0b01'
-> SPIR-V : input='0b00000010', output='0b01000000'
(Assuming my understanding on the LLVM intrinsic and the SPIR-V instruction is correct)

Shouldn't this LLVM code be converted to:
```
%tmp = OpBitReverse %input
%output = OpShiftRightLogical %tmp (8 - llvmInt.bitWidth())
```
```

https://github.com/llvm/llvm-project/pull/94219