Better detection of GPU/GPU modules + workaround for now

[Zorlin]

Right now we've just added automatic GPU installation on the lotusd node (currently "lotus_primary" in the code).

This means that if you have some nodes that have GPUs (such as lotus miner, worker) and a lotusd that does not have or use a GPU, you will not be able to use this playbook without a workaround.

That workaround:

• Create a host_vars directory, and a file within it named your-lotusd-machine-hostname.yml
• Inside that file, add the following:

lotus_nvidia_skip_reboot: true
install_nvidia_driver: false

• Re-run the playbook

Longer term, we should move the logic for detecting if you have a GPU out of fil_lotus, or make it so it will not try to find a GPU if you do not have your host in the host groups for lotus_worker or lotus_miner or something... actually, that sounds pretty good :)