Optimize index read logic

[birdstorm]

Current index read contains some logic that is confusing, e.g., covering index and double read use the same RegionExecTask, they could be better distinguished since they share a same children HandleExec.

Also, the logic in index read could be optimized by partitioning the handles according to region ids beforehand, so that it should reduce the cost of retrying when any RegionError occurs.

Some tasks to be done:

• [ ] refine index read logic to separate covering index from index read.
• [ ] optimize index read to partition handles by region ids.
• [ ] eliminate confusing logic in double read logic and make explain result more meaningful.

@zhexuany @marsishandsome PTAL

[zhexuany]

This issue is referenced to https://github.com/pingcap/tispark/pull/1195