[feature] Implementation of several micro-optimizations

[fako1024]

The following high-level changes are incorporated in this PR (see also the issue #83 for details / numbers):

• Switch packet capture to IP layer / zero-copy mode (less data to process, minimization of copies / allocations)
• Removal of GPPacket layer (instead direct interaction with the flow map, saves overhead and simplifies code)
• Removal ob obsolete select {} statement in core processing routine (this has been bothering me from the beginning - with the new two-way lock concept it's simply not necessary anymore)
• Several tiny improvements (like compiler hints and a removal of a few allocations in the inner loops where it might count)

@els0r I'll add a few comments in the code as usual for brevity, feel free to ping me if any further questions arise.

Final sidenote: I even tried to replace the whole map[string]*GPFlow logic with the new hashmap - it looked promising at first, but it turns out that using *GPFlow pointers as map values (and then changing counters on the pointers directly) significantly outweighs the improvements that come from the hashmap. As soon as there are more than a few items in the map the additional lookups required to change the values (even in place) bring me on par with the current implementation at best (while being less readable), so I decided to skip that.

[els0r]

Me gusta.

Would be great if you can do the rename of GPFlow to Flow

[fako1024]

Me gusta.

Would be great if you can do the rename of GPFlow to Flow

All done, thanks for the feedback!