I tested this on CSR1KVs running 03.14.02 and the config got accepted and was working. That is, BFD is up and both routers know that their peers are Graceful Restart capable.
R1#sh ip bgp nei | i Grace|BF
BFD is configured. BFD peer is Up. Using BFD to detect fast fallover (single-hop).
Graceful Restart Capability: advertised and received
Graceful-Restart is enabled, restart-time 120 seconds, stalepath-time 360 seconds
However, this isn't recommended since running BFD per neighbor and running Graceful Restart for the whole process is counterintuitive. In platforms that support NSF, Graceful Restart will allow traffic to still be forwarded via an interface or link that was detected by BFD as down which isn't desirable. If you really require Graceful Restart, what you can do is configure it on a per neighbor basis via neighbor x.x.x.x ha-mode graceful-restart so that the capability won't be exchanged on peers you're running BFD with.