Fix the arithmetic for wait_until_done(); add start_time kwarg in seconds, instead of ns.

[NHDaly]

Correctly account for the time sent from the API (ms since epoch), and
add a new parameter, start_time, which is in seconds not nanoseconds.

Deprecate the old parameter, to be removed in a future release.

Also adds an complex mocked test for wait_until_done, which tests that it really is waiting for the expected time, based on the created_on field from the API, which comes back in milliseconds.

---

The actual src/ changes are pretty straightforward:

• change from ns to seconds: the API returns the unix timestamp in ms and julia's time() function returns timestamp in seconds (to ~microsecond precision).

But then i made the PR more complicated by:

• keeping the old interface to make the change backwards compatible, and
• adding a complicated mocked unit test to make sure we're testing the behavior here and it's working as we expected

... sorry for the extra complexity.

[NHDaly]

Wow, creating this unit test was a real pain, but i think it's pretty solid!

It turned out we did already have some integration tests, so i don't understand why those weren't failing before @nantiamak's PR #103... 🤔 Very strange.

But anyway, i think this should be good once and for all.

[nantiamak]

@NHDaly Thanks for the PR! LGTM.

Just a comment about the description of the PR: "julia's time() function returns unix timestamp in ms." I think time() returns seconds. You are using it for seconds in the code, so I believe the code works as expected. It's just the description that needs updating, if I understand correctly.

[NHDaly]

Thanks you're 100% correct! I think my understanding was wrong when i started, and then i fixed it in subsequent commits. Thanks very much for the review! 😊 i'll adjust that then merge.