Fix mac val cast problem

[ep-12221]

Task Description

Static cast long long overflow.

Solution Description

Enhance range checking.

Passed Regressions

core-test

Upgrade Compatibility

Other Information

Release Note

[ep-12221]

The mapping Dima issue is about the precision conversion problem in Mac seekdb's small farm.

[hnwyllmm]

[P2] In round_floating_to_uint64, the checks for rounded <= INT64_MIN_AS_DOUBLE and rounded >= UINT64_UPPER_BOUND_AS_DOUBLE are handled before checking is_column_convert. In non-strict mode, OB_DATA_OUT_OF_RANGE is downgraded to a warning, but the current out_val is still persisted. This leads to incorrect results for extreme values in unsigned columns: a DOUBLE value of -2e28 gets written as 2^63 (MySQL should clamp it to 0), and a DOUBLE value of +2e28 gets written as INT64_MAX (MySQL should clamp it to UINT64_MAX; the FLOAT path writes it as 2^63). MySQL's Field_longlong::store(double) handles negative numbers and upper bounds according to unsigned column semantics first: https://github.com/mysql/mysql-server/blob/8.0/sql/field.cc. It is suggested to move the column-convert branch earlier, so negative numbers uniformly output 0 and values reaching 2^64 output UINT64_MAX. The non-column-convert branch should retain the existing differences between FLOAT/DOUBLE CAST AS UNSIGNED. Also, add test cases for FLOAT/DOUBLE ±2e28 -> BIGINT UNSIGNED column conversion, as the current unit tests do not cover these two boundaries.