Add hash algorithms to pna.p4

[hesingh]

Currently hash algorithms are TODO in PNA specification. This PR attempts to complete the TODO and also add defines to use as a flag to hash api to use hash on only source, destination, or both IP/IPv6 addresses, and protocol.

One has to start somewhere and thus psa.p4 hash algorithms have been copied to pna.p4. Also, none of the hash algorithms look like one a Nic can't support.

[jafingerhut]

Looks reasonable to me. If a particular NIC cannot support some of these, it is pretty common practice for particular targets to only support the subset that they can, and I expect that to be true of PNA target devices as well.

[hesingh]

Please merge, thanks.

[jfingerh]

@hesingh Happy to do so after having at least an announcement and call for objections or concerns in the next public PNA work group meeting next Monday. I do not anticipate any, but I'm trying not to ram things through without such notification.

[thantry]

Would be good to add Toeplitz and Xor hash, which are common in many NICs for RSS.

[thomascalvert-xlnx]

LGTM.

Agree that Toeplitz needs adding, but equally happy for that to be in a separate commit.

[jfingerh]

@thantry @thomascalvert-xlnx Thanks for review. Yes, we should add other hash algorithms as desired. For Toeplitz, this issue is tracking that, and from previous conversations it sounds like there might be additional control plane and/or data plane APIs desired there, and if so, merely adding a new enum value to the list is not enough, and I'd rather someone propose all of the pieces necessary in a PR addressing that issue: #19

For XOR, I have created a separate issue to track that: #93 I understand the desire to pile all the desired changes into single PRs, but hopefully we can get to a place where separate PRs can be moved through the pipeline of writing, approvals, and merging quickly enough that we are comfortable doing separate PRs.