Pre-create Packet Tracer containers

[gomezgoiri]

One of the main problems of creating Packet Tracer instances (i.e., Docker containers running PT) on demand it that it takes a long time to start them (see preliminary performance testing results).

This forces users to wait up to 15 seconds before start interacting with their widget. Furthermore, it makes unit testing unnecessarily slow.

An improvement would be to create containers before users demand them and then pause these containers (so they won't consume CPU, only memory) before they are needed.

[gomezgoiri]

A first iteration could pre-create containers as a response to API calls (issue #7).

A more sophisticated algorithm should take into account the following metrics to decide whether the server should pre-create more containers or stop doing it:

• How many containers are being used right now? How many available containers are unused? Checking this, we can somehow predict if more containers will be needed soon.
• The current CPU (Set a CPU limit to avoid overloading servers with PT containers #8), memory (Set a memory limit to avoid overloading servers with PT containers #9) and disk usage (Set a disk limit to avoid overloading servers with PT containers #10). If the server is overloaded it does not make sense to keep creating containers. Instead of this, clients should get an unavailability error.