Fix issues caused by scanning ranges which don't start with a text node

The rects returned by range.getClientRects() could include the entire start element's bounding box.

ext/fg/js/document.js

@@ -195,6 +195,11 @@ function docSentenceExtract(source, extent) {
 }
 
 function isPointInRange(x, y, range) {
+    // Require a text node to start
+    if (range.startContainer.nodeType !== Node.TEXT_NODE) {
+        return false;
+    }
+
     // Scan forward
     const nodePre = range.endContainer;
     const offsetPre = range.endOffset;