Use gbm device from image copy protocol instead of default node

This fixes `BufferConstraints` errors for workspace capture on Nvidia
outputs of a Intel+Nvidia system. (That didn't happen on Amd+Nvidia;
maybe a linear stride alignment issue?)

src/backend/wayland/mod.rs

@@ -27,13 +27,15 @@ use cosmic::{
     },
 };
 use futures_channel::mpsc;
-use std::{cell::RefCell, collections::HashMap, fs, path::PathBuf, sync::Arc, thread};
+use std::{cell::RefCell, collections::HashMap, sync::Arc, thread};
 
 mod buffer;
 use buffer::Buffer;
 mod capture;
 use capture::Capture;
 mod dmabuf;
+mod gbm_devices;
+use gbm_devices::GbmDevices;
 mod screencopy;
 use screencopy::{ScreencopySession, SessionData};
 mod toplevel;
@@ -59,7 +61,7 @@ pub struct AppData {
     capture_filter: CaptureFilter,
     captures: RefCell<HashMap<CaptureSource, Arc<Capture>>>,
     dmabuf_feedback: Option<DmabufFeedback>,
-    gbm: Option<(PathBuf, gbm::Device<fs::File>)>,
+    gbm_devices: GbmDevices,
     thread_pool: futures_executor::ThreadPool,
 }
 
@@ -310,7 +312,7 @@ fn start(conn: Connection) -> mpsc::Receiver<Event> {
             capture_filter: CaptureFilter::default(),
             captures: RefCell::new(HashMap::new()),
             dmabuf_feedback: None,
-            gbm: None,
+            gbm_devices: GbmDevices::default(),
             thread_pool,
         };