Add --samples to hirm binary

[ThomasColthurst]

This PR is not ready to check in yet, because actually using --samples is way too slow in practice. For example, the integration test in this PR

/bazel-bin/hirm --seed=1 --iters=5 assets/animals.unary --samples=5

currently hangs for over ten minutes.

However, I believe that is a problem in the current implementation of HIRM::sample_and_incorporate rather than anything this PR adds.

If/when this PR is submitted, it will address #178

[ThomasColthurst]

Updated with code to
(1) Populate the domain_crps with the already observed entity id's before sampling and
(2) Unincorporate any entities that end up not being used because r_contains_items is true.

Even with these updates, though, the code still doesn't generate samples in a reasonable amount of time. The reason appears to be somesort of mismatch between the entity id's observed by the irm->domains->crp.assignments and the entity id's observed by get_relation(r)->get_data(). For example, when running

./bazel-bin/hirm --seed=1 --iters=5 assets/animals.unary --samples=5

I see only entity id's 0 through 49 being added to domain animal, and that is all irm->domains->crp.assignments sees as well, but get_relation(r)->get_data() contains entity id 50.

[ThomasColthurst]

Besides the fix to sampling, this commit also updates write_observations to be able to handle entity id's that were created during sampling. I went with the boring "new_[DomainName]_[EntityId]" naming scheme for these; let me know if you have a better idea.

[emilyfertig]

Thanks!