fix(hardware): multiple nvidia driver installation fixes

[sigreturn]

This pull request fixes a couple of issues related to the installation of the NVIDIA driver:

1. The DKMS variant of the driver was being used even when it wasn't necessary. It will now only be installed if you've selected a non-standard kernel package such as linux-zen or linux-lts, which should help speed up the installation process for a majority of users.
2. It brings the installation process in line with the packaging changes announced yesterday

[h8d13]

I'll just ping these here again: #3828 #3830

If I understand correctly, there is also an option that doesn't need dkms or headers correct (but would only work for the mainline kernel?). EDIT I see you already working on that (::

We might want to actually merge this early as proprietary is already a dead option (seen 2-3 reports of error logs already on reddit). Then, work on enhancements, just to avoid issues incoming 🤔

About Sway disclaimer, I always thought it was unnecessary (I've used sway on Nvidia countless times?)

Also we should move kernel selection before profile (graphics drivers). perhaps after bootloader would make for better logic.

[sigreturn]

This should be enough for now, I tested it briefly on my machines and it seems to work fine, so I'm marking this ready for review. I had (and still have) other changes planned, but I'm quite busy due to the holiday season rapidly approaching so I'm putting those on hold for now so we can at least get something working in time for the monthly .iso release next week.

I'm also not too sure about the locale changes I made in 416235a, I'm assuming they are fine and it's up to the translators to update that for their respective language? If that's not how this works, feel free to let me know and I can revert that commit or bring it in line with how things are supposed to be. :)

[svartkanin]

I need to look into this more as it will take quite some testing to make sure this doesn't break anything.

In any case though the changes are not backwards compatible. Unfortunately, we're currently saving the selection value to the config

which isn't ideal when parsing the config. It would be better to have that changed to the key instead which is some more work but needs to be taken care off either in this change or a future PR.

[sigreturn]

Ouch, didn't think about that... that's quite an annoying problem to deal with because it seems no matter how we go about this, there will be at least some breakage.

I'll have to think about this more in depth after the holidays, but in the moment I feel like this is one of those situations where breakage is possibly warranted and better in the long run. If we make the config start using keys instead of the selection's label and just accept the breakage, it would make dealing with potential problems like this much easier.