Make sure InMemorySessionManager stores session before notifying listeners

[manovotn]

I came across this when re-visiting quarkusio/quarkus#46363 and I believe this is the correct fix for it.

That being said, I am not sure what other parts of Quarkus (or user code?) might possibly rely on the current ordering of listener notification versus session storage.
@gsmet any idea what should I check? Obviously I tried the Undertow tests in Quarkus repo but that feels a little insufficient.

Also cc @mkouba as you were at least vaguely aware of the original issue :)

[quarkus-bot]

Status for workflow Build

This is the status report for running Build on commit 3745984.

Failing Jobs

⚠️ Artifacts of the workflow run were not available thus the report misses some details.

Status	Name	Step	Failures	Logs	Raw logs
✖	Build - JDK 17		⚠️ Check →	Logs	Raw logs
✔️	Build - JDK 21			Logs	Raw logs

[manovotn]

The CI failure is very unlikely to be related:

The hosted runner: GitHub Actions 45 lost communication with the server.

Although it did fail again on rerun 🤷

[manovotn]

Even with a release, you are lightning fast @gsmet, thank you!
I'll send a PR to Quarkus tomorrow along with the test case I had for the original issue :)