KVM is just like ESXi when using as transport nodes in that NSX-T requires access to dedicated, physical NICs. You would want two for failover purposes, and KVM only supports the teaming policy of Failover Order. You can optionally use a LAG (which is still dedicated pNICs), but on KVM with LACP you can only have one.
Thanks for the answer daphnissov.
In the NSX-T reference design
Figure "6-8 KVM Compute Rack Failover Teaming" depicts non overlay traffic (for example, a storage network) added to the KVM N-VDS. But gives no further details on how to do so. While I can imagine different ways to do it, I am left wondering if it is ok for a non NSX-T entity to fiddle with the configuration of the N-VDS, which is actually a NSX-T managed OVS instance, where the pnics exist in a NSX-T controlled bridge.