Add GDB* support to Numba generated GPU kernels.

[diptorupd]

Currently, Numba-DPPY has not debugger support for GPU kernels. This feature request tracks the tasks that should be completed to enable debugger support in Numba-DPPY.

• Evaluate what DWARF/debug symbols dpcpp generates for kernels.
• Add flags to Numba-DPPY compilation to generate the same debug symbols for kernels as DPCPP when user has specified "debug=True".
• Work with application debugger team (Matti Puputti) to identify steps on how to run Numba with GDB*
• Develop examples to demonstrate how the debugger should be used.

[diptorupd]

Useful links

https://numba.readthedocs.io/en/stable/developer/debugging.html?highlight=GDB
https://numba.readthedocs.io/en/stable/user/troubleshoot.html?highlight=GDB#debugging-jit-compiled-code-with-gdb
https://numba.readthedocs.io/en/stable/user/troubleshoot.html?highlight=GDB#debugging-jit-compiled-code-with-gdb

[diptorupd]

@akharche I experimented with a bit today and managed to get the DWARF generated. I manually set the debuginfo compiler flag (compiler.Flags().set('debuginfo')) and had to comment out:
https://github.com/IntelPython/numba/blob/de35af55d295f1677cca76646691d8c51c79d3cf/numba/dppl/target.py#L258 that adds an alwaysinline attribute to the spir_func we generate.

After these two steps I was able to get the !dbg symbols generated. I think the first step will be to enable the debug=True flag for compile_with_dppl and make sure if the flag is set we add debuginfo and also do not add the alwasyinline.

[akharche]

Extended by #149

[PokhodenkoSA]

@diptorupd What examples did you mean in checkbox "Develop examples to demonstrate how the debugger should be used"?