When using WLR_BACKENDS=headless with WLR_DRM_DEVICES, the renderer
would ignore the device selection and pick the first available render
node. This is problematic for multi-GPU setups where a specific GPU
should be used for rendering.
Parse WLR_DRM_DEVICES (colon-separated) in open_drm_render_node() to
allow selecting which DRM device to use, matching the behavior of the
DRM backend. The specified path can be any node type (card, render,
primary) and the corresponding render node will be opened.