WoW Custom Metrics

[klshuster]

Patch description
Add a few custom evaluations for wizard of wikipedia:

1. When doing the knowledge selection task, the teacher will compute additional title/checked sentence recall metrics; the latter measures whether the checked sentence appears in the text of any of the returned candidates
2. When doing the response generation task, the teacher will compute f1 score of the generation w.r.t. the checked sentence

We skip the other test now because... it's not super necessary

Testing steps
Added tests to CI.

$ pytest tests/tasks/test_wizard_of_wikipedia.py
=====test session starts =====
platform linux -- Python 3.7.9, pytest-6.2.1, py-1.10.0, pluggy-1.0.0.dev0
rootdir: /private/home/kshuster/ParlAI, configfile: pytest.ini
plugins: hydra-core-1.0.0, requests-mock-1.8.0, regressions-2.1.1, datadir-1.3.1
collected 2 items

tests/tasks/test_wizard_of_wikipedia.py ..                                                                                                                                                                                                                                                                                                                           [100%]

=====slowest 10 durations =====
1.81s call     tests/tasks/test_wizard_of_wikipedia.py::TestWoW::test_custom_eval

(5 durations < 0.005s hidden.  Use -vv to show these durations.)
====2 passed in 3.65s =====

[stephenroller]

Whooo love when people use custom_eval