Add way to get index of thread in its threadpool

[IanButterworth]

A proposal to make it easier to index a thread within a thread pool.

i.e. take the following that breaks if there's an interactive threadpool because both nthreads() and @spawn default to the :default pool, which comes second.

x = zeros(nthreads())
Threads.@spawn x[Threads.threadid()] = 1

This introduces

x = zeros(nthreads())
Threads.@spawn x[Threads.threadpool_threadidx()] = 1

Needs consideration for whether this helps the bigger picture of avoiding task migration sensitive design issues.

[Seelengrab]

In my opinion, this won't really help the general issue. Similar to how PHP now has three different functions for escaping SQL, without a hard removal people still use the existing functionality. No amount of blog posts or warnings in docstrings is going to (meaningfully) change that, because old examples/APIs have a large amount of inertia.

People are already used to this (dangerous) pattern from other languages, so using threadid in spite of its problems is more or less second nature. Not to mention that a name like threadpool_threadidx is a lot harder to figure out as the "obviously correct way of doing things", since it's 1) longer 2) non-obvious (you need to know about julia threadpools already to even consider it as a potential solution/tab completion) and 3) not obviously making the bad pattern safer.

IMO, we should accept that the API Base.Threads exposes in its entirety, together with the names it uses for everything, is detrimental to the underlying task-based concurrency story.