Fix inability to interrupt processes on Windows

[itamarst]

Fixes #3400

(or at least fixes the Windows subprocesses variant; there's a bunch of stuff in there that is probably separate tickets—at least pdb was moved out into its own).

It appears that in addition to wait() being uninterruptible (https://bugs.python.org/issue28168), so is read()ing from the subprocess.

Reading is therefore pushed off to threads, and wait() replaced with the equivalent poll()-based loop. If the user interrupts the process, it is killed.

This is demonstrated by a unit test that previously failed, and now passes. Still remaining for me to do is actually test this fix on Windows from Jupyter; I'm waiting for a VM to unpack. If a reviewer wants to try this out themselves, by all means do so :)

[itamarst]

I now have a process interrupting test that simulates how ipykernel handles interrupts, and it demonstrates that system()-run proceses:

1. Exit quickly on Linux in response to the interrupt.
2. Ignore the interrupt on Windows.

[itamarst]

It appears the subprocess also becomes uninterruptible while reading, not just during wait().

[itamarst]

Investigating the Travis failure.

[Carreau]

Well you seem to know what you are doing and have a windows machine. I can't test, but the code seem reasonable. Thanks !

[amrita112]

Should this be working in IPython version 7.15.0? I couldn't get it to work.
I ran the following in a jupyter notebook cell:
input()
and then tried to interrupt the kernel, but couldn't.

[itamarst]

@amrita112 you also need the newest IPykernel.