Added RPL-lite, a lightweight implementation of RPL

[simonduq]

Addresses #8

• Re-design of the overall RPL architecture;
• Removed support for storing mode, multi-dag, multi-instances;
• Clear cut between packet parsing/framing and processing;
• Clear cut between processing/state update and resulting control traffic;
• Closer to RFC6550 when it comes to DIO timer and parent management;
• All functions are documented with Doxygen;
• New test for DAO-ACK;
• And many other improvements...

This PR makes RPL-lite the new default. It can be configured out from makefile with CONTIKI_WITH_RPL_LITE=0

[simonduq]

Does anybody have a clue how to fix this https://travis-ci.com/sics-iot/contiki/jobs/79744807
gpg: no ultimately trusted keys found

[simonduq]

The problem is now gone, was a server-side thing I guess

[simonduq]

Addressed Niclas' issue: now using full include path (in core).
Rebased and added support for logging.

[nvt]

Niclas, are you reviewing this or should I?

[simonduq]

please wait, there is more coming

[simonduq]

The PR is now ready. As a side effect of making sure this is reliable in a real setup, I had to include a couple changes not directly related to RPL: mostly on the logging module, TSCH, and link-stats.

Some of the added features since last time:

• Removed config flags that define network size and density, as suggested by @nfi
• Added support for leaving a DAG after route poisoning
• After joining, nodes will not select a parent immediately. Instead, they wait to have a parent that is fresh & acceptable
• Nodes now start advertising their DAG and rank only after they have become reachable, i.e., in the DAO-ACK case (now default), only after receiving a positive DAO-ACK. In practice, nodes first hear a DIO and initialize their DAG from it, then they probe some neighbors and eventually select one, then send a DAO, and when getting a DAO-ACK start to advertise, in turn. Takes about 15-20s. This was discussed with @joakimeriksson
• Added time-hysteresis to MRHOF, so that nodes now disregard the rank hysteresis for a candidate parent that has been a consistently better option over a given time span (10 minutes by default).

This is tested in the SICS internal testbed on 25 firefly nodes, with the ipv6/rpl-simple example, with bidirectional traffic between any node and the root. Works both on top of CSMA and TSCH. With TSCH, end-to-end reliability is not sampled with statistical significance yet but seems beyond 99.9%.

[joakimeriksson]

I see no problem merging this as it is not taking away support for the old-style Contiki RPL with storing mode (for anyone who still needs that).