pageserver: batch get page requests and serve them with one vectored get

[VladLazar]

We don't take advantage of queue depth generated by the compute
on the pageserver. We can process get page requests more efficiently
by batching them.

Hold get page requests for configurable maximum debounce timeout in order
to facilitate merging. Then process the entire batch via one get_vectored timeline operation.
By default, no merging takes place.

Sub-Tasks / Punted

Beta Give feedback

1. https://github.com/neondatabase/cloud/issues/20620
2. small Retroactive RFC, reuse write-ups in PRs, they're quite good
3. page_service: figure out correctness of holding the TimelineHandle in the pending batch #9850
   +0
4. page_service: batching needless waits for unbatchable requests #9835
   +0
5. page_service: unit-test batching logic #9834
   +0
6. page_service: measure tail latency impact in batchable workload #9837
   +0

Implementation

Beta Give feedback

1. feat(page_service): timeout-based batching of requests #9321
   +0
2. page_service: add benchmark for batching #9820
   +0
3. page_service: getpage batching: refactor & minor fixes #9792
   +0
4. WIP: page_service: higher-resolution timer for batching #9822
   +0
5. page_service: rewrite batching to work without a timeout, pipeline in protocol handler instead #9851
   run-benchmarks +1
6. WIP: batching observability improvements #9870
   run-no-ci +1