Fix race condition during Python library loading

[andrewwhitehead]

An error could be triggered when the library is loaded from multiple threads simultaneously (for example if a thread pool is used). This update adds a lock around the library initialization.

The cached_property dependency is also removed.

[swcurran]

Can you provide some details on where this was found and so who might be interested in upgrading sooner than later?

[andrewwhitehead]

This was found in did-webvh-py when verifying proofs on multiple threads. I haven't seen it previously.

[ff137]

This is a massive fix.

We had an issue with connecting many concurrent users to an issuer (using didexchange, via the issuer's public did).

If there were too many simultaneous requests, the automated flows just never completed, without any errors or exceptions being raised. I was investigating last week, and it seemed to potentially be related to many steps in acapy opening and closing an askar session, in short succession. But I really couldn't tell what was wrong.

We just tested with using askar 0.4.2 in our acapy fork, and the issue is resolved 🎉 thanks so much for finding/fixing it!