Cuda 12 and related patches

[tomsmeding]

Here are a bunch of changes that should make cuda ready for a new release, in my view. What do you think? Any critique welcome. (As I said before, a release of 0.12.8.0 does not preclude a release of 0.12.8.1 soon after.)

I ended up reverting some of the DeviceResources values that I said before I'd keep, mainly because in the commit that introduced the differing values, the chip name still had a (?) after it. :P

Closes #74.
Closes #75.
Fixes #73.
Fixes #77.

[tmcdonell]

I did not read this extensively but I trust your judgement. You might consider (maybe as a separate PR) to just drop the entire runtime module: it has not been updated nor tested in ages, and does not fit in well with how to use CUDA from Haskell, so it's mostly just a headache to keep around at this point. It should really be its own package.

(some other libraries, e.g. cublas, technically assume/require the runtime interface, but they are easy to make work with the driver interface anyway)

[tomsmeding]

Given that cuda 13 is already out, let's at least get this out.

[tomsmeding]

What I forgot to say: @noahmartinwilliams thank you for the various PRs, including over at language-c, helping fixing the state of the accelerate build :)