fix: guard against null rect in Node.set rect

[OddballGreg]

Fixes the unhandled TypeError: rect is null reported in #540.

The bug

Node.set rect lets a WINDOW node hold a null rect, because that case breaks immediately and never touches an actor. split() then copies that value onto a freshly created CON node:

// tree.js:1120
newConNode.rect = node.rect;    // node is a WINDOW whose _rect is still null

The CON branch does dereference it, so set_size() throws:

JS ERROR: TypeError: rect is null
  set rect@.../lib/extension/tree.js:129:11
  split@.../lib/extension/tree.js:1120:5
  command@.../lib/extension/window.js:567:19
  trackWindow@.../lib/extension/window.js:1418:16
  trackCurrentWindows@.../lib/extension/window.js:1651:12
  _bindSignals/this._displaySignals<@.../lib/extension/window.js:211:18

Reached when auto-split-enabled is true (the default), since trackWindow issues a Split command before the new window has been through a renderTree pass and therefore before it has any geometry.

This matters more than a log line because the throw escapes a display workareas-changed handler, which runs while mutter is reconfiguring monitors and work areas. See #540 for the display-side symptoms I measured alongside it.

The change

-        if (this.actor) {
+        if (this.actor && rect) {

Guards the actor geometry calls rather than the assignment in split(), on the grounds that any caller can reach the setter with a null rect, so the setter is the right place to be defensive.

The node still records _rect = null. Only the two actor geometry calls are skipped, and the following renderTree assigns a real rect and sizes the actor, so there is no visual change.

Verification

• Running this locally on GNOME Shell 46.0 / Wayland with a 3 monitor layout that reproduced the exception three times in one session. The exception is gone and tiling behaves normally.
• Formatting checked against the pinned prettier@2 from package.json: the file is clean both before and after the change. (Note prettier@3 reports pre-existing issues in this file, unrelated to this PR, so I left them alone rather than bloat the diff.)
• node --check passes.