Add two variants of PNA

[shirshyad]

High Level Comments

• To me based on IPSEC requirement which requires packet parsing once before decrypt action and once again post decrypt action, two architectures are being proposed. I am not sure this approach of architecture variants based on offload functionality will set a precedence that can lead to more architectures in the future either because offloads are implemented uniquely by various vendors or due to constraints of tooling. If a common abstraction is not possible to arrive at, then should we leave IPSEC offload out of standardization effort and let vendor specific extern/s express IPSEC at the cost of portability ?
• It would help to hear from the end user of P4 who would use NIC from multiple vendors. We should consider their inputs before we checkin two version of PNA architectures.
• If there is conclusion that more than one architectures is unavoidable, then should we just describe abstract model that merges both PNA1 and PNA2 and state in the specification that every vendor need not implement all the blocks of the abstract model and hence call out that portability is impacted.

[jafingerhut]

I plan to close this PR as soon as I determine that there is nothing I want to salvage from it. It uses an approach that I have given up pursuing for inclusion in PNA.

[jfingerh]

I have looked through the changes in this PR, and did not find any changes that were unrelated to the 2-parser/2-control vs. 1-parser/1-control idea, so closing this PR as abandoned.