statistics: add concurrency for async merge global stats to improve performance

## Enhancement
### Background
In the new TiDB version, `async merge global stats` feature is supported to reduce memory usage and merge stats time.

Async merge global stats is really important for our business clusters, because more than 100+ partitioned tables exist and each partition is partitioned by day range. The largest number of partition now is about 2200+.

But in my sysbench test, merging 4096 partition stats takes about **8 mins** and it brings some risks for online.

### Optimization
parallel merge fmsketch/cmsketch/histogramAndTopN, so that the performance of merging 4096 partition stats improves about 30%.