I have been facing some gunicorn issues while fixing tickets. The gunicorn fails start because the pid has been taken by some other process.
This bug has been solved in the v19.7.0.

Almsost a year ago Nicolas fixed the version of gunicorn to v19.4.5 here 391b026e
Since then the bug stoping up from upgrading has been fixed as well.

Without logs of SlapOS it is hard to understand the problem in case of
failure, thus add part of this information into log file.

Add a Mosh (mobile shell) component https://mosh.org/ , especially for the "local echo" feature ( see [demo](https://youtu.be/XsIxNYl0oyU?t=6m11s)) that makes remote shell over high latency network less painful.

---
I could not get the integrated `mosh` command to work with webrunner's openssh, but I'm already opening this MR for visibility.
For now, connecting using the way described as **Q: How do I run the mosh client and server separately?** in mosh [FAQ](https://mosh.org/#faq) works.

You can evaluate this with:
```ini
[buildout]
extends = component/mosh/buildout.cfg
parts = mosh
```

Start the server from the server side (after ssh'ing to the server) :
```bash
$ LC_ALL=C.UTF-8 ./parts/mosh/bin/mosh-server new -s -v
```
it will prints something like:
```
MOSH CONNECT 60002 XXXeu5nj+CLZtw/HiIcrLQ

mosh-server (mosh 1.3.0) [build mosh 1.3.0]
Copyright 2012 Keith Winstein <mosh-devel@mit.edu>
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>.
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.

[mosh-server detached, pid = 24525]
```

then from client connect with:
 ```bash
$ MOSH_KEY=XXXeu5nj+CLZtw/HiIcrLQ mosh-client 2001:X:X:X:X  60002
```

/reviewed-on nexedi/slapos!164

When a file is removed while rsync is running (after build the list and before transfer the file), rsync fail with the follow message:

     rsync warning: some files vanished before they could be transferred

Long back rafael proped a solution here 387d3b7e
But it seems like adding --ignore-missing-args dont really help but just ignore the files given in args and are counldn't be found anymore.

/reviewed-on nexedi/slapos!225

   This library is required for a generic implementation of OAuth implementation of facebook login

/reviewed-on nexedi/slapos!223

numpy  >= 1.13.1 is required for numpy.core.multiarray

Since nexedi/erp5@6c6b77a0, this is now a safe
operation, in that a mistake while moving databases won't result in a loss of
the catalog if the zodb is empty.

This new implementation relies on
nexedi/erp5@6bb4c56e, which means that zopes
decide on their own to create the site at startup if there's none, and the
'erp5-bootstrap' service is removed from the root partition.

/reviewed-on nexedi/slapos!217

Keep url in publish parameters, as it is used by other requests.

  There is no reason to git clone by default, as slapos.cookbook should be developed independently.

  Anyone an include develop, locally, to test on the edge changes, or include local modification
  when they use webrunner or a development server.

-  add --nouse-image flags to cros_sdk to prevent usage of missing lvm2
-  add dynamic discovery of ebuild files instead of rely on hard-coded versions

/reviewed-on nexedi/slapos!219

/reviewed-on nexedi/slapos!220

JupyterLab is provided side-by-side with Jupyter Classic interface with same process.

Published URL changed:

 * jupyterlab-url provides IPv6 URL to JupterLab interface
 * jupyter-class-url provides IPv6 URL to the classic UI

Thanks to using only one process, the path in the URL differentiates the style of the interface.

Note: This MR replaces closed nexedi/slapos!207

/reviewed-on nexedi/slapos!218

component/gnutls: version up libgcrypt 1.8.1, including security fixes. also version up gnutls 3.5.15.