Do not require float-time's arg to fit in time_t (Bug#11825).

Do not require float-time's arg to fit in time_t (Bug#11825).

This works better on hosts where time_t is unsigned, and where
float-time is applied to the (negative) difference between two times.
* editfns.c (decode_time_components): Last arg is now double *,
not int *, and means to store all the result as a double, without
worrying about whether the seconds part fits in time_t.
All callers changed.
(lisp_time_argument): Remove last int * arg, as it's no longer needed.
All callers changed.
(Ffloat_time): Do not fail merely because the specified time falls
outside of time_t range.