Replace imitation environments with seals

[Rocamonde]

Description

Fixes #501. (Depends on HumanCompatibleAI/seals#58).

[AdamGleave]

Note will need to remove mentions of envs from the docs (especially once #525 is merged)

[codecov]

Codecov Report

Merging #541 (95d37ca) into master (531fa06) will increase coverage by 0.03%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #541      +/-   ##
==========================================
+ Coverage   97.45%   97.48%   +0.03%     
==========================================
  Files          88       83       -5     
  Lines        8401     8079     -322     
==========================================
- Hits         8187     7876     -311     
+ Misses        214      203      -11     

Impacted Files	Coverage Δ
src/imitation/algorithms/mce_irl.py	95.95% <100.00%> (ø)
tests/algorithms/test_mce_irl.py	100.00% <100.00%> (ø)

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[Rocamonde]

Tests are passing now. The errors were quite silly:

• The first was that the base POMDP was incrementing the number of steps after the done bool had been computed, so it was always running one step over the horizon.
• The second was that in the notebook examples, I was passing the same environment instance to each parallel vector environment instead of creating a new one.

I think that the seals pull (HumanCompatibleAI/seals#58) should be good to merge, except perhaps for adding some tests. @AdamGleave do you have any suggestions as to what tests we could add? Once we merge that, we replace the version pointer on this PR and we can merge this too.

[AdamGleave]

Mostly looks fine -- satisfying how much code we were able to delete :) Definitely was the right move to migrate to seals.

I'm not sure about the from seals import base_envs as envs, OTOH would prefer it to just be from seals import base_envs as envs seems sufficiently generic it could easily cause clashes down the line and is another name to keep track of (what's envs? oh, it's seals.base_envs) but I'm happy to hear arguments to the contrary.

Other suggestion is to beef up the seals testing with some of the code removed in this PR. Obviously not something to resolve in this PR, but I'd feel better if we either decide against that or open an issue/PR in seals to track that before we close this PR and forget about it.

[AdamGleave]

LGTM