* Add decoding benchmark

* Add .gitattributes

* Add test data

* Implement decoding speed improvements

* Performance improvements

* Fix tests

* Optimizing values vs ptr.

* Change cache inserts to pointers

* Refactor tests for ip dedup changes

* Fix tests

* Fix types in device protocol tests

* Fix tests

* Fix types in routingtable tests

* Fix kernel package for darwin/win

* Swap BTree with Map

* Drop legacy BMP code

* Drop legacy fib implementation.

* Fix Ptr calls

* Drop unused code

* Drop unused Less functions

* Drop unused copy function

* Drop dead code. Add IP dedup to prefix dedup.

Co-authored-by: Marcus Weiner <marcus.weiner@gmail.com>

* Replace ip/prefix map caches with btree caches.

* Cleanup

* Fix magic number

* fixed build

* Add caches

* Replace IPs and Prefixes with pointers to these

* Fix nil pointer derefs

* Fix nil pointer deref

* Fix community encoding

* Fix tests

* Fix tests

* Fix tests

* Fix tests

* Fix tests

* Fix tests

* Fix BGP path hashing

* Fix path attr decoding

* Fix tests

* Fix tests

* Fix tests

* Fix tests

* Cleanup

* Cleanup

* Revert config change

* Fix Benchmark. Fix ClusterList PA.

* Fix ClusterList tests

* Add client

* Remove stale test

* Cleanup

* Increase verbosity on decode failure

* Improve chache layer performance

* Improve bgp path cache performance

* Cleanup

This reverts commit a5bccc29, reversing
changes made to 88c1c813.

With this commit i'm introducing the netlink protocol to read routes
from the linux kernel routing table to the locRIB.
This also adds the functionality to insert new routes to the linux
kernel routing stack which where learned by other protocol (like e.g.
BGP).

This is done by the protocols/netlink/netlink.go object.
The netlink.go file acts as an facade around the netlink_reader.go and
netlink_writer.go.

netlink_reader itself reads in a defined reding interval new routes from
the kernel and propagates new routes to all subscribed clients.
The netlink reader can also be filtered using the same filter-api as in
adjRibIn/adjRibOut for BGP.
To read new routes to the locRIB you have to register the locRIB to the
netlink_reader using the ClientManager interface, just like you would
register the locRIB to the adjRibIn.

netlink_writer must be subscribed to the locRIB in order to write the
routes from locRIB into the kernel.
This useses the ClientManager interface, just like you would register
the adjRibOut to the locRIB.
netlink_writer itself currently does not support filters, since all
routes form the locRIB should be written to the kernel.
The filter functionality could be easily added here as well.

The netlink_writer currently does not support ecmp. This is subject of
an other commit.

The current implementation could be considered as work in progress, but
the code is mature enough to start a review on it.