Preload indexes on followers

[kungasc]

In 24-1 we removed passing of FlatIndex pages from leaders to followers and it seems to be a cause of high tail latency on followers

To resolve this we should:

• Reproduce that behaviour on a slice
• Remove extra disk reads on followers during transaction execution process
  • Fully preload Sys, UserTables, Snapshots tables
  • Preload indexes
• Remove passing any other pages from leader to followers (should be loaded during a part switch)

Connected to YDBREQUESTS-3667, KIKIMR-18751

[kungasc]

Experiment: 1 thread read from a follower with simultaneous writes

Timings:

• 00-10 - 1 writer, batch size 1
• 10-20 - 10 writers, batch size 1
• 20-30 - 50 writers, batch size 1
• 30-40 - 50 writers, batch size 100

ydb-stable-23-4-11

ydb-stable-24-1-14-hotfix-7

[kungasc]

Improved latency in database: