Document the risk of actions getting replayed in FixedUpdate

[cBournhonesque]

If we have a system in FixedUpdate that calls `buffer_message_to_send, then during each rollback the system would get called multiple times again, which means the message would be sent multiple times.

Maybe we should provide a schedule FixedUpdateWithoutRollback for actions that should happen in FixedUpdate but shouldn't be rolled back?

[msvbg]

I don't think more schedules should be added, since there's already a lot of schedules in Bevy and adding more is both confusing and reduces available parallelism. Maybe buffer_message_to_send could detect that Lightyear is in rollback and no-op automatically, returning an error value.

[cBournhonesque]

That's a good idea!