Bump LLVM tarballs on 1.4

[staticfloat]

This rebuilds and upgrades the LLVM tarballs on 1.4, backporting some naming changes and such from 1.5, but functionally should be identical.

[staticfloat]

Hmmm, the win32 tester has failed the Distributed and SuiteSparse tests pretty regularly, despite my attempts to rebuild it. I'm going to keep trying, but if I can't get it to go through I'm going to call this more than just bad luck.

[KristofferC]

I also reran it once with those failures.

[staticfloat]

Okay, it's happened enough times now that I think it's a real failure.

We're getting EXCEPTION_ACCESS_VIOLATION during the Distributed test suite pretty reliably, it seems to be happening in this test:

julia/stdlib/Distributed/test/distributed_exec.jl

Lines 762 to 779 in 4f718ae

	# issue #13168
	function f13168(n)
	val = 0
	for i = 1:n
	val += sum(rand(n, n)^2)
	end
	return val
	end
	let t = schedule(@task f13168(100))
	@test t.state == :runnable
	@test t.queue !== nothing
	@test_throws ErrorException schedule(t)
	yield()
	@test t.state == :done
	@test t.queue === nothing
	@test_throws ErrorException schedule(t)
	@test isa(fetch(t), Float64)
	end

We're getting another EXCEPTION_ACCESS_VIOLATION in the SuiteSparse tests here:

julia/stdlib/SuiteSparse/test/threads.jl

Lines 3 to 20 in 4f718ae

	@testset "threaded SuiteSparse tests" begin
	A = sprandn(200, 200, 0.2)
	b = rand(200)
	function test(n::Integer)
	_A = A[1:n, 1:n]
	_b = b[1:n]
	x = qr(_A) \ _b
	return norm(x)
	end
	res_threads = zeros(100)
	Threads.@threads for i in 1:100
	res_threads[i] = test(i + 100)
	end
	@test res_threads ≈ [test(i + 100) for i in 1:100]
	end

It looks threading related to me; I don't understand why this is failing now.

[staticfloat]

I've been unable to narrow this down any further. I vote we release as-is.

[KristofferC]

Okay. Let's merge this, rerun CI on my backport branch, merge that if things look OK and release.