compute: metrics for stuck/failing getpage requests to alert on pageserver unavailability

[jcsp]

We should have a general alert for computes to report if they can't reach a pageserver for too long. To catch issues like #10309, where the pageserver is hung internally and it's not obvious exactly what's wrong & which endpoint is affected, without knowing which clients couldn't get what they wanted.

Ideas:

• metric for the longest-waiting request currently in flight
• metrics for ratio of failed/succeeded requests in recent history

@ololobus we might need your team's help/advice to build this