Cythonize device memory resource attributes

[leofang]

Description

There are a few pure Python constructions that are getting in the way, and those also forced us to use a single value type (was int and being changed to uint64_t in #1272). We have LLM now so we really should just spell it out once and tweak them quickly manually, like what we do for device attributes and program options.

Before (main):

In [6]: %timeit mr.attributes.reuse_allow_internal_dependencies
265 ns ± 6.03 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

After this PR:

In [6]: %timeit mr.attributes.reuse_allow_internal_dependencies
116 ns ± 0.63 ns per loop (mean ± std. dev. of 7 runs, 10,000,000 loops each)

By construction this also fixes https://github.com/NVIDIA/cuda-python-private/issues/197.

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[leofang]

/ok to test 74c6611

[rwgk]

Oh, this is awesome! Please disregard my comment on #1272. I didn't see this before; and I didn't look around enough to realize it can be done this elegantly.

[Andy-Jost]

Thanks for putting this together, Leo. This looks like the right approach to me.

[leofang]

Thanks guys! Merging.

[github-actions]

Doc Preview CI
Preview removed because the pull request was closed or merged.