Coalesce non-vectored get with vectored code (use vectored get for single key gets)

[jcsp]

Anticipate some performance work to make vectored get equally fast for single key case (e.g. #6932)

Ordered list of changes

Preview Give feedback

1. pageserver: publish the same metrics from both read paths #7486
   +0
2. pageserver: finish vectored get early #7490
   +0
3. pageserver: coalesce read paths #7477
   +0
4. pagserver: use vectored read path in benchmarks #7498
   run-benchmarks +1

[jcsp]

Last week:

• this shook out a bug that was fixed.
• investigated why it's slower in some cases.

This week:

• A fix for difference in how we handled aux files
• Continue optimizing

[VladLazar]

Last week:

• Merged aux file handling fix
• Subtle fix for materialized page cache usage
• Optimised
• Benchmarking - looks on par with legacy read path (even a bit better)
• Ran with validation in staging over the weekend - no correctness issues

This week:

• Run in staging without validation to qualify perf
• Start staged roll out if above is ok

[VladLazar]

Last week:

• Qualified perf & correctness in staging/pre-prod

This week:

• Deploy to ap-southeast regions & monitor

[VladLazar]

Last week:

• Deployed to ap-southeast-*. Small latency increased, but tput was much higher. CPU utilisation was better

This week:

• Deploy to us-east-*

[koivunej]

#7939 should be reverted when the default is changed.

[VladLazar]

Rollout was completed a few weeks ago. At this point, the non-vectored read path is no longer in use (apart from validation which will be removed by #8005)