Refine building ranges logic

[xuyifangreeneyes]

Enhancement

Currently the logic of building ranges has following problems:

1. No memory control. We can construct the case like Query crashed the Tidb server #30755 to let TiDB OOM by building too many ranges.
2. Building ranges is expensive. When there is a long IN list in query, it takes lots of CPU and generates lots of temporary small objects, which leads to frequency GC.
3. The logic of building ranges is complex and not easy to read. Maybe we can refactor it step by step.

memory control for building ranges

• *: control memory when building ranges #37160
• sessionctx/variable: add system variable tidb_opt_range_max_size #37706
• *: restrict column range mem usage #37714
• *: restrict index range mem usage #37754
• planner, util/ranger: restrict mem usage for index join inner ranges #38129
• sessionctx/variable: put TiDBOptRangeMaxSize to right place and set default value to 0 #37835
• util/ranger: use different full range for pk, not-null col and normal col when range fallback happens #37769
• sessionctx/variable: turn on range memory control by default #38592
• session: set tidb_opt_range_max_size to 0 when upgrading to v6.4.0+ #38694
• add system variable tidb_opt_range_max_size docs-cn#11239

reduce cost for building ranges

• util/ranger: avoid building ranges twice in detachCNFCondAndBuildRangeForIndex #37177

refactor for building ranges

• planner: reuse DetachCondsAndBuildRanges logic for both index and cluster index #37178
• util/ranger: reuse convertPoint and validInterval logic #37831
• planner: small refactor for index join inner range building #38100