Support different state size and fix exception in venv.__del__

[Trinkle23897]

close #351 and #346

I add a try...except ValueError block because of the following reason:

• if the observation is a dict/other class: np.stack automatically converts to dtype=np.object
• if the observation is a normal array (with the same shape): np.stack behaves normally as previous
• if the observation is a list of array with different shape or classes like nx.Graph which rewrite __getitem__: we need to manually convert to np.object
• if the observation is a list of list (e.g. np.array([[1], [2,3], [4,5,6]])): np.stack automatically converts this type of data to np.object
• other things such as rew/done/info: their type are pre-determined, i.e., rew is a float, done is a boolean, info is a dict, and they can be handled by np.stack correctly. No need to change these logics.

Also there's workaround related to #147: do not raise error when it finds list of np.array with different shape[0].

As for #346, __del__ causes several bugs even before this issue, so just simply remove it and let the user decide whether or not to close the env instead of automatically closing in the end.

[codecov-commenter]

Codecov Report

Merging #352 (2c1cc1d) into master (bbc3c3e) will increase coverage by 0.15%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #352      +/-   ##
==========================================
+ Coverage   94.30%   94.46%   +0.15%     
==========================================
  Files          53       53              
  Lines        3406     3413       +7     
==========================================
+ Hits         3212     3224      +12     
+ Misses        194      189       -5     

Flag	Coverage Δ
unittests	94.46% <100.00%> (+0.15%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
tianshou/data/batch.py	99.74% <100.00%> (+<0.01%)	⬆️
tianshou/env/venvs.py	91.66% <100.00%> (+0.36%)	⬆️
tianshou/policy/modelfree/trpo.py	93.33% <100.00%> (-0.11%)	⬇️
tianshou/data/collector.py	95.35% <0.00%> (+0.42%)	⬆️
tianshou/env/worker/subproc.py	93.95% <0.00%> (+2.68%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update bbc3c3e...2c1cc1d. Read the comment docs.