I have option 2 working. Like you, I found it a better fit for the existing network design in situations where there is an existing robust MPLS WAN. The key is corresponding conditional routes on the MX at both ends.
- MX A has a route to site B VLANS through MX B if it can ping MX B
- MX B has a route to site A VLANS (or 0.0.0.0) through MX A if it can ping MX A
- If the ping fails, the VPN also has that route, but with a lower priority, so the VPN takes over
Most of my sites already had OSPF in place, so most of these are actually "While host responds" not "While next hop responds" and they are pinging the OSPF gateway at the remote site. The hub site injects the default route into OSPF and the remote sites have a high metric default route to the MX, so the OSPF default takes priority. If the MPLS circuit fails, OSPF loses that route and the default goes to the MX, where the VPN takes over. At one site, the backup is via cable modem, not cell, so it will also advertise the default route, but with a worse metric than the hub site. Our MPLS is over a point-to-multipoint, so any sites that can get to that site can share that cable modem.
My preference is to have a Cradlepoint or other LTE gateway device (like the Meraki MG if you want to stay on brand) connected to the WAN, rather than the built-in USB LTE backup. It works much better and the MX doesn't complain. If I don't have that, I have the WAN also connected back through a separate VLAN on the MPLS circuit so the MX can get to the dashboard, but I don't route any traffic that way.
This let me add the MX for LTE/cable modem backup without disrupting my existing network.