Regenerate bindings for NVML 12.9.79. Update device tests for new NVML_FI_DEV fields

[blthayer]

Latest v12 bindings (12.9.79). Note this is a breaking change. See also: #134 and #135

[blthayer]

Hey folks, a quick update: I'm not planning to pursue this much further, as instead I need CUDA 13 support (which has some backwards incompatible changes). But I'll leave this open for now in case it's useful. See also: #134

[blthayer]

Note the current v12 fields in main don't match the latest NVML v12 header...

[TheElectronWill]

It's a pity that this is a breaking change…
Something should be done in nvml-wrapper to support both versions IMO, otherwise it will be very brittle. When we use nvml-wrapper, we don't know which version of NVML will be used at runtime.

[blthayer]

It's a pity that this is a breaking change… Something should be done in nvml-wrapper to support both versions IMO, otherwise it will be very brittle. When we use nvml-wrapper, we don't know which version of NVML will be used at runtime.

@TheElectronWill - thanks for weighing in. I agree, this incompatibility should probably be handled via a runtime version check and match statement or something handled by this wrapper. That being said, the CUDA versioning is a little weird/complex and the breaking change occurred between versions of CUDA 13.0 (see #134). As far as I can tell (although I would need to dig deeper), some v13 field numbers slipped into this wrapper's v12 bindings about a year ago?

I'm not a maintainer of this repo, just opened a couple draft PRs to help illustrate the differences and give folks branches to use if they want.

[blthayer]

FYI, the maintainers are working on a solution: #137

[blthayer]

#137