Lightning import throughput should be balanced across all TiKVs when the table is pre-split

[db-will]

Enhancement

When using Lightning to import data, if pre-split is performed on the table and node rate limiting is enabled on TiKV, it cause continuous range regions to be distributed on the same store. This results in store-level hotspots during the import process, causing the import speed to fall far short of expectations. Currently, PD only determines balance and scatter regions based on the number of regions per store. Therefore, even if Lightning calls scatter regions before data import, it does not have the desired effect.