Tablet splitting

[mbautin]

Jira Link: DB-735
Enables dynamically re-sharding the data in a cluster. This is accomplished by allowing each shard (called a tablet) in the cluster to split into two or more tablets at runtime, transparent to the end user.

Prerequisites

✅ Design Doc: automatic re-sharding of data with tablet splitting

Phase 1 - Basic functionality

✅ #1458 Splitting DocDB data files
✅ #1459 Support splitting on the nodes (YB-TServers)
✅ #1460 Refresh tablets seamlessly across nodes
✅ #1461 Drive tablet splitting from master
✅ Ability to manually split tablets (hash and range)

Phase 2 - Automatic splitting

✅ #1461 Mechanism to initiate tablet splitting automatically (from YB-Master leader)
✅ Auto-splitting for range-partitioned tables
✅ Auto-splitting for hash-partitioned tables
✅ #1462 Size-based tablet splitting strategy
✅ #6101 Policy on when to stop splitting
✅ #4929 Cleanup parent tablets from disk after split

Phase 3 - Advanced splitting strategies

✅ #5523 #6424 Compaction improvements to limit the disk utilization
⬜️ #1463 Load-based tablet splitting strategy
⬜️ #5854 Handle YCQL/YSQL retries internally so we avoid application errors
⬜️ #5373 Integration with 2DC
⬜️ #6667 Limit the number of outstanding splits from the master