Revise default sizes for list, array, string, bytes, ... generators

[jmid]

As part of discussing "The Great Renaming" ™️ offline, @c-cube and I agreed to revise the default sizes used in our
list, array, string, bytes, ... generators.

Currently, these tend to use nat which has a terrible distribution:

qcheck/test/core/QCheck_expect_test.expected.ocaml5.64

Lines 1513 to 1536 in 1aa167f

	+++ Stats for nat dist ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
	stats dist:
	num: 1000, avg: 382.03, stddev: 1290.55, median 9, min 0, max 9890
	0.. 494: ####################################################### 850
	495.. 989: ###### 93
	990..1484: 8
	1485..1979: 3
	1980..2474: 7
	2475..2969: 1
	2970..3464: 3
	3465..3959: 3
	3960..4454: 3
	4455..4949: 4
	4950..5444: 3
	5445..5939: 3
	5940..6434: 3
	6435..6929: 1
	6930..7424: 2
	7425..7919: 1
	7920..8414: 1
	8415..8909: 5
	8910..9404: 3
	9405..9899: 3

Before the 1.0.0 release we should revise these to have a nicely curved (and more predictable) distribution, e.g., based on exponential/float_exp with a reasonable mean parameter:

qcheck/test/core/QCheck_expect_test.expected.ocaml5.64

Lines 1828 to 1851 in 1aa167f

	+++ Stats for exponential 10. dist ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
	stats dist:
	num: 5000, avg: 9.45, stddev: 9.82, median 6, min 0, max 79
	0.. 3: ####################################################### 1631
	4.. 7: ##################################### 1109
	8..11: ######################### 747
	12..15: ################# 512
	16..19: ########## 308
	20..23: ######## 251
	24..27: ##### 165
	28..31: ## 85
	32..35: ## 64
	36..39: # 43
	40..43: 25
	44..47: 20
	48..51: 20
	52..55: 3
	56..59: 8
	60..63: 3
	64..67: 1
	68..71: 2
	72..75: 2
	76..79: 1

In the qcheck-STM library we already do so (adding a bit of skew to avoid generating too many identical 0-length lists):
https://github.com/ocaml-multicore/multicoretests/blob/b5be1fe48bef28a1bbe604d59b322918624f987f/lib/STM.ml#L175-L180