[try_put_and_wait] Test coverage should be significantly extended

[kboyarinov]

          I see that all of the tests are ultimately completely serial. The function_node is serial but also the flow graph executes in an arena with a single slot. I suppose this is done to make the results deterministic since most of the tests are whitebox tests that know the exact processing order. Is there risk that serial execution could hide some bugs caused by concurrency?

Originally posted by @vossmjp in #1412 (comment)