Fix sporadic failures testing with JRuby (#1012)

* Wait for blockers to proceed

If this spec proceeds to reset the barrier before there are any
waiters, the reset will not trigger them and the spec will fail.
This at least tries to ensure that the waiter is there and will
be interrupted as expected.

* Bump join timeout up for slower system

On slower or heavily-loaded systems, such as CI, this join may
need more than one second. Give it ten before giving up and
raising an error.

* Use repeat_until_success to wait for waiters

This avoids getting stuck and waiting forever if something goes
awry.

* Also skip on JRuby until we can fix sporadic failures

spec/concurrent/atomic/cyclic_barrier_spec.rb

@@ -71,6 +71,7 @@ module Concurrent
         end
 
         start_latch.wait(1)
+        repeat_until_success { expect(barrier.number_waiting).to eq 1 }
         barrier.reset
 
         expect(barrier).not_to be_broken

spec/concurrent/edge/erlang_actor_spec.rb

@@ -918,7 +918,7 @@
         end
 
         specify "timing out" do
-          skip('flaky on truffleruby') if Concurrent.on_truffleruby?
+          skip('flaky on truffleruby and jruby') if Concurrent.on_truffleruby? || Concurrent.on_jruby?
 
           count_down = Concurrent::CountDownLatch.new
           body = { on_thread: -> { m = receive; count_down.wait; reply m },

spec/spec_helper.rb

@@ -53,7 +53,7 @@ def requires=(paths)
   config.after :each do
     while defined?(@created_threads) && @created_threads && (thread = (@created_threads.pop(true) rescue nil))
       thread.kill
-      thread_join = thread.join(1.0)
+      thread_join = thread.join(10.0)
       expect(thread_join).not_to be_nil, thread.inspect
     end
   end