Decide whether or not to support the target kwarg in the numba.jit decorator

[hameerabbasi]

In a couple of issues, e.g. #5868, users have reported that using the target= kwarg for the numba.jit decorator doesn't always work. While this isn't documented on the website (website docs), it is documented in the docstring, and so anyone running numba.njit? in ipython or Jupyter will find it. An excerpt from the docstring is below:

    target: str
        Specifies the target platform to compile for. Valid targets are cpu,
        gpu, npyufunc, and cuda. Defaults to cpu.

So, there is a decision to be made. Does the Numba team support the kwarg (which means writing tests, fixing up bugs etc.), or not (which means deprecation and removal).

[stuartarchibald]

Thanks for raising this. I think e.g. 'gpu' is not a target, the rest are valid as there are function registries for these and we should see if they working/if it makes sense for the code path to exist.