Supervisor: Retry on "Address already in use", and/or implement JEP 66

[jmcphers]

There is a race condition inherent in the Jupyter protocol; because there is a gap between the time ZeroMQ socket ports are selected by the client and the time they are bound by the server, another process can bind to the ports before the server does. The result is a ZeroMQ error that looks like this when the server tries to bind to the port:

zmq.error.ZMQError: Address already in use

Here's a screenshot of this error happening in Positron:

This race condition should be rare, and we have already taken steps to mitigate it -- for example, the kernel supervisor already keeps track of ports that are "reserved" by kernels that have not started yet. However, there is still a small chance that any kernel startup will result in this error, and the chance is higher during automated tests since a lot of startups happen quickly.

To address this, we could:

• Implement the JEP 66 Kernel Handshaking Pattern. This is already implemented in ark and would just need implementations in the supervisor (and IPYKernel?). And/or:
• Have the supervisor recognize this error (possibly just by scraping the output) and retry the connection automatically instead of reporting a start failure.

[isabelizimm]

Noting that I just hit this error after a "normal" refresh of the Python interpreter in

Positron Version: 2025.01.0 (Universal) build 87
Code - OSS Version: 1.95.0
Commit: 240e51fa165f6c512e586f0af105a0c8fc092607
Date: 2024-12-16T02:50:00.442Z
Electron: 32.2.1
Chromium: 128.0.6613.186
Node.js: 20.18.0
V8: 12.8.374.38-electron.0
OS: Darwin arm64 24.1.0