Ok, I found a better workaround, since the previous one doesn't work super well. This should make things work okay for now, hopefully without any (too many) ill effects, at least ones that aren't worse than part of the cloud just dying.
Compatible with previous build, but update recommended so people can see you properly online.
Workarounds:
- Swapped to a modified version of managed WebSocket client with disabled Utf8 validation and switched SignalR back to websockets
-- This is only temporary workaround to this problem, since it technically breaks WebSocket spec and can lead to undefined behavior
-- I've tested this for a bit and it doesn't seem to have any noticeable ill effects - at least not worse than what was already happening
-- Once I manage to narrow down the exact cause, the validation will be restored