[SYCL-MLIR]: Add convert-spirv-builtin-to-nvvm Pass

[SealSoul117]

Enable SYCL-MLIR to Run on NVIDIA GPUs
This patch introduces a new pass to support executing SYCL-MLIR-generated code on NVIDIA GPUs.
The pass is intended to be applied to the final MLIR after lowering with cgeist, where the IR is already in the LLVM dialect.

The pass performs the following key transformations to enable compatibility with NVIDIA backends:

1. Mark kernel functions so they can be recognized as GPU kernels.
2. Replace SPIR-V-specific global variables and accessor logic with direct NVVM intrinsic calls, ensuring compatibility with the NVIDIA CUDA backend.

With this pass, I have successfully executed SYCL code on an NVIDIA GPU using the following steps:

1. Use cgeist to lower SYCL C++ code to MLIR in the LLVM dialect.
2. Apply this pass to the generated MLIR using sycl-mlir-opt.
3. Convert the transformed MLIR to LLVM IR using mlir-translate.
4. Integrate the resulting LLVM IR into the standard DPC++ NVIDIA compilation pipeline by replacing the corresponding IR.
5. Successfully generate a runnable binary for execution on NVIDIA GPUs.

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.