restrict bytes_until_sample to int64_t. This allows optimal asm

generation of sub bytes_until_sample, usize; je; for x86 arch.
Subtraction is unconditional, and only flags are checked for the jump,
no extra compare is necessary.  This also reduces register pressure.

src/prof.c

@@ -1168,6 +1168,9 @@ prof_sample_threshold_update(prof_tdata_t *tdata) {
 	uint64_t bytes_until_sample = (uint64_t)(log(u) /
 	    log(1.0 - (1.0 / (double)((uint64_t)1U << lg_prof_sample))))
 	    + (uint64_t)1U;
+	if (bytes_until_sample > SSIZE_MAX) {
+		bytes_until_sample = SSIZE_MAX;
+	}
 	tsd_bytes_until_sample_set(tsd_fetch(), bytes_until_sample);
 
 #endif