fix undefined float-to-integer cast at the numeric limit boundary

[aysha-afrah26]

The safe-cast helpers in convert_impl.hpp and safe_any.hpp gate a floating point to integer conversion with from > static_cast<From>(numeric_limits<To>::max()), and if that passes they static_cast the value to the integer type. The trouble is that when the target has more significant bits than the source mantissa, like double to int64_t or float to int32_t, static_cast<From>(To::max) rounds up to a value that is one past the real maximum: for int64_t it becomes 2^63, which is not representable in an int64_t. A double that lands exactly on 2^63 therefore slips past the > check and reaches static_cast<int64_t>(2^63), which is signed-overflow undefined behavior. I hit it with UBSan while poking at numeric ports, since this path is reachable from a script bitwise operator on a double and from Blackboard::set on a port already locked to int64_t (isCastingSafe/ValidCast). The fix keeps the same limits but switches the upper bound to a half-open comparison when static_cast(To::max) rounds up, decided at compile time from the digit counts, so the boundary value is rejected before the cast instead of after. Smaller targets like int32_t whose max is exactly representable keep the inclusive > and still accept their true maximum. I applied the same change at both sites and added a regression test that trips UBSan before the fix and passes after.