The issue is due to the controller being unaware of all the VTEPs configured on the edge host, in particular the VTEP IP of the vdr-vdrport. The code path that handles TRUNK configuration only informs the controllers about one of the VTEPs and not that of the vdr-vdrPort mapped to the other VTEP.
When the host (where the VM resides) VDR ARPs for the next hop (ESG interface) over the transit VNI, it is sent only to one of the VTEP-IP/vmnicX of the Edge Host. Since the vdr-vdrPort on Edge host is mapped to a separate VTEP-IP/vmnicX, if the packet received on the wrong VTEP/uplink, it is dropped by the RPF check filter on the vswitch, resulting in ARP resolution failure , thereby disrupting S-N traffic towards the Edge.