Add test of v1model update_checksum implementation

[jafingerhut]

It does not test the verify_checksum implementation.

[jafingerhut]

The new P4 program only compiles with the new v1model checksum APIs added via this commit: f09a95d

but later reverted on 2017-Aug-28.

In my fork of the p4c repo, I created a branch from the commit just before the changes were reverted on 2017-Aug-28, but when I created a PR on Github from that branch, Github seems to be treating it as a commit at the top of master, and hence the program fails to compile because verify_checksum() and update_checksum() are not defined.

[cc10512]

Thank you for putting this together. Since it is failing on master, can you please add the test to the list on known failures (backends/bmv2/CMakeLists.txt:106 and backends/p4test/CMakeLists.txt:52)? We'll remove it when we merge back the checksum support.

[mihaibudiu]

There should already be a few tests with stf files that exercise the checksums in bmv2.
But more tests are always better.

[jafingerhut]

@mbudiu-vmw I found only one P4_16 test program in the p4c/testdata/p4_16_samples program where there is checksum calculation code, and there is an .stf file to specify input packets, and expected output packets that check for the correct checksum being computed. That is issue655-bmv2.p4 and issue655-bmv2.stf. It only calculates checksums for a single 16-bit field.

The test program and .stf file proposed in this PR calculate the IPv4 header checksum over an entire 20-byte IPv4 header, plus a test packet that has 24-bit IPv4 header with options, parsed into a varbit field with a non-0 length. Thus this test should be adding a little bit of functional coverage not already covered by other tests.

[jafingerhut]

@cc10512 I have tried making the changes you suggested -- thanks for file name and line number pointers to guide me. We'll see what the Travis build thinks of these changes in a bit.