Avoid unnecessary precision loss for pgbench's --rate target.

It's fairly silly to truncate the throttle_delay to integer when the only
math we ever do with it requires converting back to double.  Furthermore,
given that people are starting to complain about restrictions like only
supporting 1K client connections, I don't think we're very far away from
situations where the precision loss matters.  As the code stood, for
example, there's no difference between --rate 100001 and --rate 111111;
both get converted to throttle_delay = 9.  Somebody trying to run 100
threads and have each one dispatch around 1K TPS would find this lack of
precision rather surprising, especially since the required per-thread
delays are around 1ms, well within the timing precision of modern systems.

Author: tglsfdc

src/bin/pgbench/pgbench.c

@@ -169,7 +169,7 @@ double		sample_rate = 0.0;
  * When threads are throttled to a given rate limit, this is the target delay
  * to reach that rate in usec.  0 is the default and means no throttling.
  */
-int64		throttle_delay = 0;
+double		throttle_delay = 0;
 
 /*
  * Transactions which take longer than this limit (in usec) are counted as
@@ -826,9 +826,12 @@ getGaussianRand(TState *thread, int64 min, int64 max, double parameter)
 /*
  * random number generator: generate a value, such that the series of values
  * will approximate a Poisson distribution centered on the given value.
+ *
+ * Individual results are rounded to integers, though the center value need
+ * not be one.
  */
 static int64
-getPoissonRand(TState *thread, int64 center)
+getPoissonRand(TState *thread, double center)
 {
 	/*
 	 * Use inverse transform sampling to generate a value > 0, such that the
@@ -839,7 +842,7 @@ getPoissonRand(TState *thread, int64 center)
 	/* erand in [0, 1), uniform in (0, 1] */
 	uniform = 1.0 - pg_erand48(thread->random_state);
 
-	return (int64) (-log(uniform) * ((double) center) + 0.5);
+	return (int64) (-log(uniform) * center + 0.5);
 }
 
 /* helper function for getZipfianRand */
@@ -5114,8 +5117,8 @@ main(int argc, char **argv)
 						fprintf(stderr, "invalid rate limit: \"%s\"\n", optarg);
 						exit(1);
 					}
-					/* Invert rate limit into a time offset */
-					throttle_delay = (int64) (1000000.0 / throttle_value);
+					/* Invert rate limit into per-transaction delay in usec */
+					throttle_delay = 1000000.0 / throttle_value;
 				}
 				break;
 			case 'L':
@@ -5239,7 +5242,11 @@ main(int argc, char **argv)
 	if (nthreads > nclients)
 		nthreads = nclients;
 
-	/* compute a per thread delay */
+	/*
+	 * Convert throttle_delay to a per-thread delay time.  Note that this
+	 * might be a fractional number of usec, but that's OK, since it's just
+	 * the center of a Poisson distribution of delays.
+	 */
 	throttle_delay *= nthreads;
 
 	if (argc > optind)