[llvm] [AMDGPU] Fix wrong operand value when floating-point value is used as operand of type i16 (PR #84106)

[Matt Arsenault via llvm-commits]

================
@@ -2045,9 +2047,26 @@ bool AMDGPUOperand::isInlinableImm(MVT type) const {
       return false;
 
     if (type.getScalarSizeInBits() == 16) {
----------------
arsenm wrote:

This is redundant with the switch over the handled 16-bit types 

https://github.com/llvm/llvm-project/pull/84106