Enable SPIR-V to device caching

[ZzEeKkAa]

Add kernel reference caching for the SPIRV to device compilation. This is a critical change to make experimental kernels usable.

Should we remove comment from kernel_dispatcher.py?

# TODO: Enable caching
# Add code to enable on disk caching of a binary spirv kernel.
# Refer: https://github.com/IntelPython/numba-dpex/issues/1197

I've add simple test to check if second run does not crash XD. I've made sure that kernel reference is taking from cache by enabling logs and making sure it reached right place.

Since we can not implement caching on kernel dispatcher level due to missing information about context that is available only at runtime, I guess we can close this issue.

Closes: #1070

• Have you provided a meaningful PR description?
• 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?

[diptorupd]

Should we remove comment from kernel_dispatcher.py?

# TODO: Enable caching
# Add code to enable on disk caching of a binary spirv kernel.
# Refer: https://github.com/IntelPython/numba-dpex/issues/1197

No, that is a different issue. Numba supports ahead of time (AOT) compilation using on disk caching of compiled binaries. That way if you run a function with cache=True flag the function gets serialized to disk and the next time Numba is started the on disk cached function is de-serialized and used. In our case, that will mean we serialize the SPIRV binary generated for a kernel to disk and next time de-serialize and use it. KernelBundle caching comes after that and unrelated.

[diptorupd]

Thank you! One step closer to making the experimental kernel really usable. :)