[clang] [llvm] [AMDGPU] Split OPERAND_REG_IMM_INT64 into signed and unsigned variants (PR #186575)

[Jay Foad via cfe-commits]

jayfoad wrote:

> Split OPERAND_REG_IMM_INT64 into two operand types to properly handle signed vs unsigned 64-bit immediate encoding:
> 
> * OPERAND_REG_IMM_INT64: Signed 64-bit integer operand
>   (uses isInt<32>)
> * OPERAND_REG_IMM_B64: Unsigned 64-bit integer operand
>   (uses isUInt<32>)

This is just bikeshedding but I would _much_ prefer calling them `OPERAND_REG_IMM_I64` and `OPERAND_REG_IMM_U64` to highlight the signed/unsigned difference and to match the I64 and U64 suffixes on instruction names.

https://github.com/llvm/llvm-project/pull/186575