[WIP] use new KernelClient.execute_interactive

[minrk]

in execute preprocessor

Significantly simplifies the preprocessor, relying on upstream implementation.

WIP because it's not released yet.

I hit two missing features that we should either drop here or add there:

1. When we interrupt-on-timeout, we need to flush outputs,
   which suggests that flushing outputs with the hook
   should be a method on its own.
2. ExecutePreprocessor allows a separate timeout for no output,
   not just a total timeout for the whole execution.

cf jupyter/jupyter_client#185

[mpacer]

@minrk is this still active?

[minrk]

I'd like to come back to it at some point, but it's far from urgent. Feel free to close if it's in the way.

[takluyver]

@minrk I believe the released version of jupyter_client has execute_interactive, if you want to come back to this.

[minrk]

I sure do! Trying to finish up some JupyterHub stuff right now, but I hope to find time for this at some point.

[akhmerov]

Is this PR still relevant?

[SylvainCorlay]

@minrk wow, that is indeed removing a lot a code!

Is this something you would like to rebase and include in the next release?

[MSeal]

Closing this issue as the code in question has been entirely rewritten and refactored in 5.5 and I don't believe this is relevant anymore. Reopen with rebase if there's still interest in something I missed :)