make CI config more robust to network

[bertsky]

Currently CircliCI fails in deploy job due to network connectivity problems during pip install. Dockerfile already has --timeout=3000 among the options for pip. Still, we get:

Connection broken: ConnectionResetError(104, 'Connection reset by peer')

Should we increase the timeout? Or is this just a bad day?

[kba]

Tough to do something sensible about something so unpredictable. Maybe something like make foo || (sleep 60;echo "Failed once, waiting 60 seconds, trying again"; make foo) to just repeat steps? Ugly as hell but if the network is just gone for more then a few seconds it might help.

[kba]

I restarted the build. @stweil @bertsky Do you have the rights to do that yourself at https://circleci.com/gh/OCR-D/workflows/ocrd_all ?

[stweil]

I don't have the rights.

[bertsky]

fixed as of #77