There are some services that are done on the primary PAN such as acting as a root CA for ISE internal CA deployment, serving ISE sponsor portal backend redirection, management, etc. However, if you have a normal ISE deployment where you have dedicated PSNs then restarting the primary node wouldn't have any impact on the authentication and authorization sessions because those sessions are managed by the PSNs, same for the guest flow which is also managed by the PSNs. So if you are planning to restart the primary or the secondary PAN, and there are no PSN personas on those nodes then no impact to the authentication and authorization sessions.
Regarding your last question, that depends on how you configured the network devices, for example, say you have 2x PSNs in your environment, and you configured RADIUS/TACACS on the switches to use both PSNs, the network device usually doesn't fall back to the second configured RADIUS/TACACS server unless the previous one in the list fails. The order of operation would be start from top go to bottom. If you want to load balance the RADIUS and TACACS traffic without introducing a real load balancer to your network, then what you can do is configuring half of your network devices with PSN1 on the top, and the other half with PSN2 on the top.
Also, please note that although ISE would prompt you to shutdown ISE application services if you issue the "reload" command without going first and shut those services manually, it is still recommended to shutdown the application services first, and then issue the "reload" command. ISE is a very sensitive product, and if you happen to reload it without shutting down its services gracefully, its database might get corrupted and it could actually cause the application services to fail post reload..
