chwd forces nvidia early KMS modules into initramfs, breaks Thunderbolt eGPU

Running chwd creates the 10-chwd.conf with MODULES+=(nvidia nvidia_modeset nvidia_uvm nvidia_drm) which casues nvidia to load during initramfs, before Thunderbolt enumerates the eGPU.

Result: GPU lost from the bus, Xid 154, no display output

Setup: RTX 3080 in ADTLINK UT3G Thunderbolt eGPU enclosure

Fix: emptying the MODULES line -- nvidia loads fine later via udev

It would be nice if chwd could detect if the nvidia GPU is behind a Thunderbolt bridge and skip adding the modules for it, or at least warn the user.

Start with chwd's generation of 10-chwd.conf, especially the MODULES line, and reproduce the RTX 3080 setup in the ADTLINK UT3G Thunderbolt enclosure if available. Check how early KMS affects initramfs enumeration and define completion as avoiding the broken configuration or warning users when the GPU is behind a Thunderbolt bridge.

Written by the indexing model from the issue text.