fix undefined double to int64 cast in the script complement operator

[aysha-afrah26]

The scripting operator ~ is evaluated in ExprUnaryArithmetic::evaluate (operators.hpp) by casting the operand to double and then to int64_t behind a hand-written range check, rv > double(INT64_MAX) || rv < double(INT64_MIN). That check lets NaN through, since every comparison with NaN is false, and it lets exactly 2^63 through, since double(INT64_MAX) rounds up to 2^63 and 2^63 is not greater than itself; both reach static_cast<int64_t>, which is undefined for values outside the range, and it also truncates fractional operands silently, so ~1.5 evaluates as ~1. The binary bitwise operators a few lines below go through the checked Any::cast<int64_t> instead and reject a non-integer, so the complement is the one remaining hand-rolled cast in the evaluator (the shared helpers it now relies on were tightened at the boundary in #1162). The operand is untrusted script text from Script nodes and _skipIf/_while/_onSuccess attributes, or a blackboard double: ~(0.0 / 0.0) and ~9223372036854775808.0 both trigger UBSan float-cast-overflow at operators.hpp:137 in the ASan+UBSan configuration the CI runs. The fix routes the complement through the same checked cast the binary operators use, so NaN, infinities, out-of-range and fractional values raise the existing RuntimeError and integral values, including both int64_t boundaries, evaluate as before. Header-only, no API or ABI change. The new test ParserTest.BitwiseComplementRejectsNonIntegers aborts under UBSan on master and passes with the change.