So here are my tests and results.
1 Spoke and Hub with a standard MTU of 1500 and 1 Spoke configured with 1400 :
Spoke A will still continu to use a MTU of 1500. The only effect that Spoke B has on the topology is that the Hub will clamp his own MSS to 1292 so traffic inbound to the hub will be clamped down to 1292 for every single spoke.
Traffic from Spoke A going to the internet ( in a split tunnel configuration ) will STILL use a MSS of 1460 ( MTU : 1500 )
My second test was to introduce a lower MTU in the path of AutoVPN. The only way that I was capable of doing so was to plug Spoke A into MX B ( which was no longer path of the AutoVPN domain ). The MSS advertised by both the Spoke A and HUB were still showing 1392 , which is not possible without fragmentation since the path MTU contains a hop with a lower MTU.
Yes ICMP fragmentation needed messages were flowing quite frequently. But I still think that I will have to lower the MTU on every single spoke instead of relying on PMTUD / icmp fragmentation needed packets.
TL;DR : Setting the MTU on the Spoke to a desired MTU seems the easiest way to account for a known Path MTU that is lower than 1500 ( eg : GRE ). However , relying on Meraki's Support to do it in the backend and loosing vision about that setting ( because it's a backend option ) worries me a lot.