* data.c: Avoid integer truncation in expressions involving floats.

* data.c: Avoid integer truncation in expressions involving floats.

* data.c: Include <intprops.h>.
(arith_driver): When there's an integer overflow in an expression
involving floating point, convert the integers to floating point
so that the resulting value does not suffer from catastrophic
integer truncation.  For example, on a 64-bit host (* 4
most-negative-fixnum 0.5) should yield about -4.6e+18, not zero.
Do not rely on undefined behavior after integer overflow.