Add job benchmark loop

[Kobzol]

This PR adds the main logic required to execute job benchmarks in the collector.

It handles:

• Updating the collector's heartbeat periodically
• Quick loading of already downloaded sysroots, to avoid redownloading them in-between jobs and also collector restarts (useful for local testing)
• Dequeing jobs, including in-progress jobs and expanding benchmark sets
• Distinguishing between transient and permanent job errors, storing job errors into the DB, and marking jobs as failed or successful
• Marking jobs that have been dequeued too many times as failed
• Reconnecting to the DB if a transient I/O/network/DB error happens, to try to refresh the DB connection

[Jamesbarford]

I could be mistaken however this reads slightly oddly; if http is not a not found 404 then we produce an error of sha not found? Then we fall through to an IO error if it was a 404 despite the sha actually not being found.

[Kobzol]

The variable name wasn't great, renamed it and added a comment.

[Jamesbarford]

We can use resp.error_for_status() so we can get the actual error message too from the response which could be useful for debugging; https://docs.rs/reqwest/latest/reqwest/struct.Response.html#method.error_for_status

[Kobzol]

error_for_status() is nice, but the reason why we don't provide more context here is simply because we have up to three (potentially different) errors. Now that we detect 404s explicitly, we could just bail out on the first non-404 error, but I'm a bit worried about backwards compatibility, I don't know if "toolchain not found" is always reported with a 404.