[mikelangmayr]

• CameraSession gains a destructor that calls disconnect_controller, closing both the primary and the control socket
• Previously nothing invoked it: there was no destructor and callers were not required to call close(), so both sockets survived until process exit and were released only once the controller noticed
• This is a resource leak fix. It does not address the controller becoming unresponsive, which was investigated separately and has a different cause: those failures occur at session teardown and still occur with both sockets closed deliberately
• The destructor swallows exceptions deliberately, since a destructor must not throw and a failed close is not worth aborting the process for
• Verified against the HISPEC tracking camera: a script that opens and exits without calling close() now logs "Archon connection terminated", and a 1000 frame autofetch run is unchanged at 854 fps with no dropped frames