Handle requests in a non-blocking fashion

[sersorrel]

This means that long-running handler methods (if they are long-running due to a long await) will not prevent further incoming requests from being handled. This means your RPC handlers can use idiomatic Python (just await the thing you're waiting for, and return the result when you have it), rather than having to return as quickly as possible and schedule some notification to be sent back to the client when the result is available.

We still wait for all handlers to complete on shutdown.

Fixes #48

[orweis]

Hi @sersorrel ,
Thank you for the submission and willingness to contribute.

I'm currently inclined to decline this PR, as #48 is not a real issue (has clear workarounds as seen in the discussion there), and risk, testing for the potential side-effects, and breaking-changes (especially for code currently assuming the current behaviour) for the changes here outweigh the benefits of being slightly more idiomatic.

At the least this change needs to be feature flagged (e.g. a variable passed to WebsocketRPCEndpoint.__init__) so code already utilizing this library won't be affected by the change.

I'll review this again later this week with more attention, to consider how and if this can be accepted.

[sersorrel]

No worries. I'm currently testing this with my own application, and am running into some slightly tricky issues (unsure whether there's a bug in the application or in this code), so probably best to leave it for the moment regardless. (this turned out to be a bug in my app)