Error with floats(max_value=<one less than inf>, allow_infinity=False, exclude_max=True) could be more helpful

[gjoseph92]

This has already been mentioned in #1859 (comment), but I couldn't find a standalone issue for it.

The error allow_infinity=False excludes min_value=inf (added in #1860 I believe) is a bit confusing to see when you are certain you didn't pass max_value=inf. It took me far longer than I'd like to realize that the highest-possible finite float was getting passed in as max_value, leaving no room for a larger value besides infinity. If worded differently, this might have saved me some time.

I assume this is happening because the min_value gets next_up'd in:

hypothesis/hypothesis-python/src/hypothesis/strategies/_internal/numbers.py

Lines 370 to 373 in 22607fe

	if min_value is not None and (
	exclude_min or (min_arg is not None and min_value < min_arg)
	):
	min_value = next_up_normal(min_value, width, assumed_allow_subnormal)

so when the check actually happens, it appears the user has passed in inf.

Even a message like allow_infinity=False exclude_max=True excludes min_value={sys.float_info.max} would have made the problem much clearer, though I think it could be worded in a more friendly way for users who are less familiar with how floats work.