PoC: profiles/nvidia: Eliminate early KMS usage in favor of proper systemd chainload

[ventureoo]

I’ve always considered early KMS as a workaround for issues with race conditions when loading between driver and display manager/graphics environment. Here is another approach, that allows you to solve this "right" in my opinion, as it simply ties graphics services to wait for NVIDIA module. To do this, we create nvidia-module-load.target and a udev rule that activates it if module loading is successful. If load failed, that’s okay, we have JobTimeoutSec, which allows nvidia-module-load.target to fail but load the rest of the services that depend on it (as Wants= implies soft dependencies).

To avoid breaking the work plymouth, we should switch to using simple-drm (that is, specify UseSimpledrm=1 in /etc/plymouth/plymouthd.conf, I thought this should be done on the calamares side). This is what Fedora does and it seems normal for them, see: https://fedoraproject.org/wiki/Changes/PlymouthUseSimpledrm

This will certainly require careful testing by other users, but from my testing, it works pretty well. The main advantage is the absence of bloated initramfs, which often becomes an obstacle for configurations with Btrfs snapshots and a separate /boot, as we now do when using limine-snapper-sync, which led us to decide to use ESP partition size 4 Gb by default, which I find excessive.

P. S. I’m still not sure if we should put it in chwd, or make it part of nvidia-utils package, because there are quite a lot of configuration files. On the other hand, I clearly don’t want to change anything in the current users' configuration to avoid any trouble.