Add a new (git) dataset generation scaffolding for tests

and use it to the generate a 'cmdbts2' test case strictly equivalent
to the CMDBTS repo.

See the swh/provenance/tests/data/README.md file for more details.

Note: this aims at making easy to write more test cases than depending
on the CMDBTS git repo on github. For example, a new test case should
come soon for situations like 'out-of-order' revisions.