plugin: parallel release fetches, faster shutdown, uninstall runner stop

- Run Plugin Manager release fetches on OS threads so they execute in

  parallel instead of serialising on the async executor.

- Reduce V4Connection shutdown poll from 3s to 100ms; the runner is killed

  and reaped without blocking host shutdown.

- Add PluginManager::remove and PluginProcess::shutdown_and_wait so

  uninstall can stop the runner before deleting files, fixing Windows

  access-denied errors on plugin removal.

- Rebuild ribbon immediately after uninstall so the tab disappears.

crates/ocs_plugin_api/src/process/manager.rs

@@ -2,6 +2,7 @@
 
 use std::path::Path;
 use std::sync::Arc;
+use std::time::Duration;
 
 use crate::host::{HostApi, HostNotification, PluginNotification};
 use crate::process::{PluginError, PluginProcess};
@@ -206,6 +207,21 @@ impl PluginManager {
         out
     }
 
+    /// Shut down and remove the plugin with `id` from the manager.
+    ///
+    /// Returns `true` if a plugin with that id was found and removed. The
+    /// process is killed and waited on so its DLL is released; this is intended
+    /// for uninstall, where the caller needs the files to become deletable on
+    /// Windows.
+    pub fn remove(&mut self, id: &str) -> bool {
+        let Some(index) = self.plugins.iter().position(|p| p.process.id() == id) else {
+            return false;
+        };
+        let plugin = self.plugins.remove(index);
+        plugin.process.shutdown_and_wait(Duration::from_secs(5));
+        true
+    }
+
     /// Begin asynchronous shutdown of every plugin process.
     ///
     /// Kills every child synchronously on the calling thread and moves the

crates/ocs_plugin_api/src/process/v4.rs

@@ -302,10 +302,12 @@ impl V4Connection {
         }
         // Dropping the writer closes the socket. The reader thread will then
         // exit and clear `alive`. Joining can deadlock if a plugin child
-        // inherited the socket handle, so we only poll briefly.
-        let deadline = Instant::now() + Duration::from_secs(3);
+        // inherited the socket handle, so we only poll briefly. The host process
+        // will kill the runner afterwards, so a long wait only delays shutdown
+        // without improving cleanup.
+        let deadline = Instant::now() + Duration::from_millis(100);
         while self.shared.alive.load(Ordering::SeqCst) && Instant::now() < deadline {
-            std::thread::sleep(Duration::from_millis(50));
+            std::thread::sleep(Duration::from_millis(10));
         }
         let _ = self.reader_handle.lock().unwrap_or_else(|e| e.into_inner()).take();
     }