Fix issues around comm and session lifecycles

[jmcphers]

This change addresses a few problems found by @DavisVaughan around session and comm lifecycles. Specifically:

• Comms were trying to close themselves on disposal even after the kernel has exited, leading to some errors logged since the close requests could not be delivered to the kernel.
• Runtime sessions dispose methods were never being called. This is an artifact from a time when runtime sessions could be started again after fully exiting. Now, "restarting" an exited runtime actually starts a new session. The dispose() method is implemented in the language pack and does things like clean up log streamers/file handles/sockets/etc.

Addresses #2908
Addresses #3964

While these changes are important for code quality and hygiene, aside from addressing some leaks, there ought to be no observable difference in behavior.

QA Notes

Since one of these issues results in errors being logged, it should be pretty easy to verify. For the other, the main risk here is that the runtime is going to get prematurely disposed -- once it's disposed we no longer get events, logs, anything (it's inert). So no heavy verification needed, but good to check start/stop/start cycles and make sure we're getting good behavior, events, and appropriate logs.

[github-actions]

E2E Tests 🚀
This PR will run tests tagged with: @:critical

^{readme  valid tags}

[seeM]

Code changes LGTM! I'll test the behavior once this comes out of draft.

[jmcphers]

@seeM should be ready for review now!

[seeM]

Looks good! I confirmed locally with the debugger that KallichoreSession.dispose and PythonRuntimeSession.dispose are being called, and there are no longer any console errors when shutting down a session.