Our e2e tests occasionally flake out with 500s

[ghostwriternr]

And specifically, they all timeout without being able to capture a sandbox to run on. We should figure out how to solve this, or at least be more resilient to this, wherein our infra receives bursty traffic and might need longer to schedule new sandboxes.

Also, what happens when a user's account limits are hit? Do we know when that happens and is that something we should even handle gracefully or let it fail so users discover it from logs?

What kind of a role can/should our Durable Object layer play in these situations?

[ghostwriternr]

We've made some improvements via #273, #296, #299 and #301 that reduce these. Will re-open an issue with specific flakiness we observe going forward.