Do not kill frontend isolate on uncaught exceptions.

[isoos]

• pub.dev is down (global outage) #4663
• uncaught exceptions are only logged
• using stderr instead of print to make these errors stand out from the rest
• if the frontend isolate exits for a reason, it still restarts it with an ever-increasing delay

[jonasfj]

Let's discuss this next week.. I see pros/cons I'm both directions.

For example: if some internal thing is failing persistently, and this causes 5xx responses, then it might be preferable to restart.

An unhandled exception at this levels typically implies undefined behavior, crashing and restarting is reliable way to recover.

Maybe we want to limit number of times we allow a restart. Like restart on first error, then avoid restarting due to unhandled errors for 5, 10, 20min, or something like that. That would degrade the service more gracefully.

[isoos]

Like restart on first error, then avoid restarting due to unhandled errors for 5, 10, 20min, or something like that. That would degrade the service more gracefully.

Done, with the frontendStarted as controlling for time. On the default service we'll have 4 isolates, so it will start by 4 min protection, then 5 minutes, ... up to 20 minutes maximum.

[jonasfj]

Just a suggestion... Overall this looks good.