This PR finally adds a way of flexibly run tests on real hardware.

Before, there was "testrunner", which could sequentially run tests on a single board, and the beginnings of a more general approach, hidden in lwip's tests.

Testrunner can now read json files (or whole trees of them) containing test and node descriptions.

Test descriptions describe things like "need 3 nodes, two of them samr21-xpro, one of them must have 6lowpan; flash gnrc_networking on the first, gnrc_border_router on the second and nanocoap_test on the third; set up the default route of the first and third to be the link local address of the second; then run these pexpect scripts; .....".

Node descriptions describe node groups of actual nodes, like "node0 is a samr21-xpro, serial number ATMLxxxx, has 6lowpan and a mma8x5x connected, node 1 is ...". The latter can be less specific if they contain a field "murdock_queue".

There are some nifty features (including files, simple templateing) in order to reduce duplication in the description files.

The testrunner can then read in a bunch of test and node descriptions, and then run tests, either every test once, or every test on each node group that can run it. It can also create command lines usable by murdocks backend DWQ.

Another commit adds initial integration into murdock.

Currently, there's only one samr21-xpro connected to my laptop, for testing.

TODO:

• produce some examples / getting-started hints
• document
• show test results in murdock output
• find safe way of hooking up nodes to CI, add (more) nodes
• split tests into long running / short running,

DO NOT MERGE! This is WIP.