simplify and speed up isdisjoint for two AbstractRange values

[nsajko]

Ranges have optimized intersect implementations, so it is fastest in general to just rely on isempty ∘ intersect.

This removes the optimization for empty ranges, however that optimization only results in worse code (extra branching) for common cases such as UnitRange{Int}. Removing the optimization for empty ranges should result in a performance regression in special cases like AbstractRange{BigInt}, when the range is empty. But I suppose that is OK.

[nsajko]

Tests fail here:

julia/test/sets.jl

Lines 481 to 518 in d3dc139

	# range fast-path
	for (truth, a, b) in (
	# Integers
	(true, 1:10, 11:20), # not overlapping
	(false, 1:10, 5:20), # partial overlap
	(false, 5:9, 1:10), # complete overlap
	# complete overlap, unequal steps
	(false, 3:6:60, 9:9:60),
	(true, 4:6:60, 9:9:60),
	(true, 0:6:12, 9:9:60),
	(false, 6:6:18, 9:9:60),
	(false, 12:6:18, 9:9:60),
	(false, 18:6:18, 9:9:60),
	(true, 1:2:3, 2:3:5),
	(true, 1:4:5, 2:1:4),
	(false, 4:12:124, 1:1:8),
	# potential overflow
	(false, 0x1:0x3:0x4, 0x4:0x3:0x4),
	(true, 0x3:0x3:0x6, 0x4:0x3:0x4),
	(false, typemax(Int8):Int8(3):typemax(Int8), typemin(Int8):Int8(3):typemax(Int8)),
	# Chars
	(true, 'a':'l', 'o':'p'), # not overlapping
	(false, 'a':'l', 'h':'p'), # partial overlap
	(false, 'a':'l', 'c':'e'), # complete overlap
	# Floats
	(true, 1.:10., 11.:20.), # not overlapping
	(false, 1.:10., 5.:20.), # partial overlap
	(false, 5.:9., 1.:10.), # complete overlap
	# Inputs that may hang
	(false, -6011687643038262928:3545293653953105048, -6446834672754204848:3271267329311042532),
	)
	@test isdisjoint(a, b) == truth
	@test isdisjoint(b, a) == truth
	@test isdisjoint(a, reverse(b)) == truth
	@test isdisjoint(reverse(a), b) == truth
	@test isdisjoint(b, reverse(a)) == truth
	@test isdisjoint(reverse(b), a) == truth
	end

However the cause, or one of the causes at least, seems to be a pre-existing bug in intersect:

• intersect(::StepRange, ::StepRange) silently overflows producing strange incorrect answers #47577