race condition in flux-wreck attach --no-follow test

[grondo]

Oops, I think I wrote an unreliable test for the flux-wreck attach --no-follow sharness test:
Obviously there is a race condition between starting the test and ensuring the output from each task has made it to the kvs before running the attach.

expecting success: 
	flux wreckrun -d -l -n4 sh -c "echo before; sleep 30; echo after" &&
	test_when_finished flux wreck kill $(last_job_id) &&
	run_timeout 5 flux wreck attach --no-follow $(last_job_id) >output.attach-n &&
	cat >expected.attach-n <<-EOF &&
	before
	before
	before
	before
	EOF
	test_cmp expected.attach-n output.attach-n

47
--- expected.attach-n	2018-04-13 17:07:35.014233718 +0000
+++ output.attach-n	2018-04-13 17:07:35.014233718 +0000
@@ -1,4 +1,2 @@
 before
 before
-before
-before
not ok 47 - wreck: attach --no-follow works

[grondo]

Hm, probably the easiest fix here is to retry flux-wreck attach --no-follow until we get all 4 expected lines. We want to not only test that --no-follow doesn't block/hang, but we do want to make sure it gets the expected lines from all tasks that already written data to their kz streams...