Fixing multithreaded tests

[udesou]

This PR aims to fix multithreaded tests as noted in #179.
To fix the issue, we needed to make sure we captured all tasks, which is done when allocating a task, and maintaining a list of all_tasks in the common ptls.

[qinsoon]

I left some reviews. However, before addressing any of them, I think we should make sure what we do in the PR is reasonable and sound. In particular, there are two questions:

1. Why are there tasks that appear in the shadow stack but do not appear in live_tasks?
2. For prev in jl_handler_t, why can't we follow it until we see a null pointer?

[qinsoon]

I saw the discussion on Slack.

Why are there tasks that appear in the shadow stack but do not appear in live_tasks?

live_tasks means tasks that are running. If a task is not scheduled, it won't appear in the live_tasks. This sounds reasonable. We probably want keep an eye on this in terms of performance. What we are doing now is possibly expensive, and we may want to maintain a list of all the tasks in the runtime.

For prev in jl_handler_t, why can't we follow it until we see a null pointer?

Though you asked, this question was not answered.

[udesou]

julia-version
JULIA_REPO=udesou/julia
JULIA_REF=fix/capture-all-tasks