Generate proper mangled name for kernel functions

[diptorupd]

• Have you provided a meaningful PR description?

Fixes how function names are generated at the LLVM IR level for numba_dpex.kernel decorated functions and auto-generated kernels for parfor nodes.

Previously an unreadable gobbledegook was generated

dpex_py_devfn__5F__5F_main_5F__5F__2E_kernel_5F_vector_5F_sum_24_1_2E_USMNdArray_28_dtype_3D_float64_2C__20_ndim_3D_1_2C__20_layout_3D_C_2C__20_address_5F_space_3D_1_2C__20_usm_5F_type_3D_device_2C__20_device_3D_level_5F_zero_3A_gpu_3A_0_2C__20_sycl_5F_queue_3D_DpctlSyclQueue_29__2E_USMNdArray_28_dtype_3D_float64_2C__20_ndim_3D_1_2C__20_layout_3D_C_2C__20_address_5F_space_3D_1_2C__20_usm_5F_type_3D_device_2C__20_device_3D_level_5F_zero_3A_gpu_3A_0_2C__20_sycl_5F_queue_3D_DpctlSyclQueue_29__2E_USMNdArray_28_dtype_3D_float64_2C__20_ndim_3D_1_2C__20_layout_3D_C_2C__20_address_5F_space_3D_1_2C__20_usm_5F_type_3D_device_2C__20_device_3D_level_5F_zero_3A_gpu_3A_0_2C__20_sycl_5F_queue_3D_DpctlSyclQueue_29_

After the fix:

_ZN8__main__28kernel_vector_sum_241dpex_fnB2v1B40c8tJTC_2fWQI8IW1CiAAYKPM6RBFDjESyhCQA_3dE10USMNdArrayIdLi1E1C14level_zero_gpuE10USMNdArrayIdLi1E1C14level_zero_gpuE10USMNdArrayIdLi1E1C14level_zero_gpuE

The new function names can also be demangled (e.g., using c++filt) into human readable format:

main::kernel_vector_sum_241dpex_fn[abi:v1][abi:c8tJTC_2fWQI8IW1CiAAYKPM6RBFDjESyhCQA_3d](USMNdArray<double, 1, C, level_zero_gpu>, USMNdArray<double, 1, C, level_zero_gpu>, USMNdArray<double, 1, C, level_zero_gpu>)

For parfor auto generated kernels:

_ZN08NumbaEnv13_3cdynamic_3e33__dpex_parfor_kernel_0_245dpex_fnB2v5B40c8tJTC_2fWQI8IW1CiAAYKPM6RBFDjESyhCQA_3dE10USMNdArrayIfLi1E1C10opencl_gpuE10USMNdArrayIfLi1E1C10opencl_gpuE10USMNdArrayIfLi1E1C10opencl_gpuE

that demangles into:

NumbaEnv::_3cdynamic_3e::__dpex_parfor_kernel_0_245dpex_fn[abi:v5][abi:c8tJTC_2fWQI8IW1CiAAYKPM6RBFDjESyhCQA_3d](USMNdArray<float, 1, C, opencl_gpu>, USMNdArray<float, 1, C, opencl_gpu>, USMNdArray<float, 1, C, opencl_gpu>)

Additionally, the LLVM module is now lightly optimized prior to spirv generation to ensure that the spir_func gets fully inlined into the spir_kernel wrapper function.

• Have you added a test, reproducer or referred to an issue with a reproducer?
• Have you tested your changes locally for CPU and GPU devices?
• Have you made sure that new changes do not introduce compiler warnings?
• If this PR is a work in progress, are you filing the PR as a draft?

Fixes #260