Fix json for header stack ref in push/pop primitives

[ChrisDodd]

Fairly trivial fix -- the only part I'm unsure of is removing the comment which just seemed to be generally wrong. While other places may handle Type_Stack specially elsewhere, they then don't look at the result of the PathExpression conversion, or prune in the preorder to avoid it completely.

[mihaibudiu]

This may be correct, but I can't figure it out just by reading. Unfortunately the way that p4-16 storage locations are mapped to metadata/headers is not very uniform - it's different for metadata, headers, stacks, and local variables. That's one reason that issue #172 hasn't been fixed. This part of the code should probably be reworked at some point.

Can you please add a test as well?

[ChrisDodd]

existing switch.p4 tests excercise this code -- its just that we don't actually run bmv2 on the resulting json, so we fail to notice that it is complaining that the json is invalid.

[mihaibudiu]

One possibility is to have an empty stf file - this causes the simple_switch to load the json and immediately exit.

[ChrisDodd]

Actually I take that back -- the bug this is fixing only manifests when you convert switch.p4 into P4_16 code with --pp and then compile that resulting P4_16 version of switch.p4. The latter process introduces copies of the header stack into temp locals prior to doing the push/pop, and it is the push/pop on the temp local header_stack that causes the problem (I think). Since we don't actually do this dump-and-reparse normally, just adding a stf won't do it.