Use Numba dispatcher stack for implementing device context

[PokhodenkoSA]

Closes #271
Contains #279

In this PR:

• Replace dpctl.device_context with numba_dppy.context.device_context. It means that users will manage context via numba_dppy instead of dpctl.
• CPUDispatcher retargets to DppyOffloadDispatcher.
• This PR works with https://github.com/IntelPython/numba/tree/wip/dispstack_alt1-patched

Problems:

• Test fail: test_with_dppy_context_target (https://github.com/IntelPython/numba-dppy/blob/374db39466a288cd22f62fe6897264f7caa080d2/numba_dppy/tests/test_with_context.py#L76)
  • When retarget to DppyOffloadDispatcher is should check that nested functions have not specified parameters like target and parallel.
• Tests fail:

FAILED numba_dppy/tests/njit_tests/dpnp/test_numpy_array_ops.py::test_unary_ops[opencl:gpu:0-sum-float32-10]
FAILED numba_dppy/tests/njit_tests/dpnp/test_numpy_array_ops.py::test_unary_ops[level0:gpu:0-sum-float32-10]
FAILED numba_dppy/tests/njit_tests/dpnp/test_numpy_transcendentals.py::test_unary_ops[opencl:gpu:0-sum-float32-100]
FAILED numba_dppy/tests/njit_tests/dpnp/test_numpy_transcendentals.py::test_unary_ops[level0:gpu:0-sum-float32-100]

Links:

• Implementation of TargetDispatcher: https://github.com/IntelPython/numba-dppy/blob/main/numba_dppy/target_dispatcher.py

[DrTodd13]

Looks good to me.

[PokhodenkoSA]

Obsolete.

[reazulhoque]

The device context was enabled in this PR: #493