Rewrite the timeouts

Change the strategy of how timeouts work, from a simple timer, to
something a bit more nuanced

crates/gitbutler-watcher/src/file_monitor.rs

@@ -11,11 +11,20 @@ use notify::{RecommendedWatcher, Watcher};
 use tokio::task;
 use tracing::Level;
 
-/// The timeout for debouncing file change events.
-/// This is used to prevent multiple events from being sent for a single file change.
+/// We will collect notifications for up to this amount of time at a very 
+/// maximum before releasing them. This duration will be hit if e.g. a build
+/// is constantly running and producing a lot of file changes, we will process
+/// them even if the build is still running.
 const DEBOUNCE_TIMEOUT: Duration = Duration::from_secs(60);
+
+// The internal rate at which the debouncer will update its state.
 const TICK_RATE: Duration = Duration::from_millis(250);
-const FLUSH_AFTER_EMPTY: u32 = 3;
+
+// The number of TICK_RATE intervals required of "dead air" (i.e. no new events 
+// arriving) before we will automatically flush pending events. This means that
+// after the disk is quiet for TICK_RATE * FLUSH_AFTER_EMPTY, we will process
+// the pending events, even if DEBOUNCE_TIMEOUT hasn't expired yet
+const FLUSH_AFTER_EMPTY: u32 = 6;
 
 /// This error is required only because `anyhow::Error` isn't implementing `std::error::Error`, and [`spawn()`]
 /// needs to wrap it into a `backoff::Error` which also has to implement the `Error` trait.