Rework handling of panning to identified features

[uclaros]

Plan A

This PR is a rework on how we jump to geometries. It was initially just a way to mitigate a GEOS crash when identifying self-overlapping multipart geometries, but quickly evolved!

Plan B

If this is too much for a last minute change, we can avoid hitting the geos bug by slightly modifying the current approach: Perform bounding box intersections instead of geometry intersections.

diff --git a/app/inpututils.cpp b/app/inpututils.cpp
index b5573eb3..02a874e8 100644
--- a/app/inpututils.cpp
+++ b/app/inpututils.cpp
@@ -330,13 +330,12 @@ QPointF InputUtils::relevantGeometryCenterToScreenCoordinates( const QgsGeometry
 
   const QgsRectangle currentExtent = mapSettings->mapSettings().visibleExtent();
 
-  // Cut the geometry to current extent
-  const QgsGeometry currentExtentAsGeom = QgsGeometry::fromRect( currentExtent );
-  const QgsGeometry intersectedGeom = geom.intersection( currentExtentAsGeom );
+  // Cut the geometry extent to current extent
+  const QgsRectangle intersectedExtent = currentExtent.intersect( geom.boundingBox() );
 
-  if ( !intersectedGeom.isEmpty() )
+  if ( !intersectedExtent.isEmpty() )
   {
-    target = QgsPoint( intersectedGeom.boundingBox().center() );
+    target = QgsPoint( intersectedExtent.center() );
   }
   else
   {

This will change the existing behavior, eg identifying C shaped geometry that is partially visible will not recenter the map canvas to the center of the visible part of the geometry, but to the visible part of the geom bbox.