Bug 1794265 - Fix infinite loop in PeekOffsetFromLine. r=dholbert

This effectively restores the blockFrame != aPos->mResultFrame check
here:

  66a061bd76 (l1.260)

This is hitting this rather odd code-path from the netscape era, which
is clearly not well tested:

  https://searchfox.org/mozilla-central/rev/ffa4d00965c5281def6d3ddcbcdf6259d38c9b9a/layout/generic/nsIFrame.cpp#8451-8482

I'd happily write a test, but I think I'd need some assistance as I
haven't been able to reproduce this with regular selection APIs / line
movement.

Differential Revision: https://phabricator.services.mozilla.com/D158926

layout/generic/nsIFrame.cpp

@@ -9004,6 +9004,11 @@ nsresult nsIFrame::PeekOffsetForLine(nsPeekOffsetStruct* aPos) {
         break;
       }
 
+      if (aPos->mResultFrame == blockFrame) {
+        // Make sure block element is not the same as the one we had before.
+        break;
+      }
+
       // we've struck another block element with selectable content!
       if (aPos->mDirection == eDirPrevious) {
         edgeCase = 1;  // far edge, search from end backwards