Update to CUDA 12.6

[jakirkham]

• Bump CUDA version to 12.6

[jakirkham]

With CUDA 12.6 am seeing the following test failure on CI:

___________________ test_duplicate_symbols_cubin_and_fatbin ____________________

device_functions_cubin = ('test_device_functions.cubin', b"\x7fELF\x02\x01\x013\x07\x00\x00\x00\x00\x00\x00\x00\x01\x00\xbe\x00x\x00\x00\x00\x0...x00\x00\x00\x00\x00\x00\x03\x00\x00\x00\t\x00\x00\x18\x80\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00")
device_functions_fatbin = ('test_device_functions.fatbin', b'P\xedU\xba\x01\x00\x10\x00\x00\t\x00\x00\x00\x00\x00\x00\x02\x00\x01\x01@\x00\x00\x...x1e\x08@\x00\x1f\x08@\x00\x00\x1f\xa4@\x00\x05\x1e\t@\x00\x1a\t@\x00P\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00')
gpu_arch_flag = '-arch=sm_70'

    def test_duplicate_symbols_cubin_and_fatbin(
        device_functions_cubin, device_functions_fatbin, gpu_arch_flag
    ):
        # This link errors because the cubin and the fatbin contain the same
        # symbols.
        nvjitlinker = NvJitLinker(gpu_arch_flag)
        name, cubin = device_functions_cubin
        nvjitlinker.add_cubin(cubin, name)
        name, fatbin = device_functions_fatbin
>       with pytest.raises(NvJitLinkError, match="NVJITLINK_ERROR_INVALID_INPUT error"):
E       Failed: DID NOT RAISE <class 'pynvjitlink.api.NvJitLinkError'>

test_pynvjitlink_api.py:90: Failed

[bdice]

Seems fine to me.

[brandon-b-miller]

CI failure seems to be from a test where the underlying lib is correctly erroring but I think there error isn't being translated into an NvJitLinkError correctly somehow. Looking into it

[jakirkham]

Fixed the package hashing error noted above: #97 (comment)

This uses a build time generated variant file as noted below

[jakirkham]

Have separated the bulk of these changes into PR: #101

As that is now in, will rebase this so it contains only the CUDA 12.6 update

[bdice]

@brandon-b-miller @gmarkall Would one of you be able to follow up on this failure?

[jakirkham]

Thanks for the assist Bradley! 🙏