It took me a VERY long time to narrow down this bug and I've confirmed it exists in 8.5.161, 8.8.130, 8.10.121, so I want to confirm it isn't a configuration issue.
Environment
Confirmed in both a production and lab environment. Vanilla Mobility Express deployment with 4 APs (2x 1852, 2x 1815). Basic L2 switch used in lab environment to confirm no VLAN or other backhaul issues. AVC under ME is basically the same as Flexconnect AVC with Local Switching, the controller just lives on the AP.
Symptoms and Issue
When AVC is enabled (even with no rules) traffic between wireless clients on separate APs is sporadicly dropped/blocked after a roam. There is no recovery even after disconnecting the client. Restarting the AP temporarily resolves the issue, but it occurs again after a roam. Traffic to other wireless clients on the same AP is not affected. Traffic to wired clients is not affected. Only traffic to other wireless clients on other APs. I originally noticed this with the Sonos iOS application on one AP attempting to communicate to a Sonos device on another AP. This drove me down an entire multicast rabbit hole, but latter confirmed the issue exist with unicast traffic, including ping. I narrowed it down to AVC by wiping the entire config, AP configs, and slowly rebuilding, testing along the way by sending pings while roaming across APs. At this point I've 100% confirmed that the issue is only present when AVC is enabled (even if there are no AVC rules) and can replicate it reliable by turning AVC on and off. P2P blocking is obviously disabled.
Thanks for the input and advice!