[ADD] Stricter checks mypy

[ravinkohli]

This PR partially addresses #230.

1. Added mypy checks for warn-redundant-casts, warn-return-any, warn-unreachable,

[nabenabe0928]

Thanks for the PR, I will put some good practices for mypy typing:

1. class rather than Dict[str, Any]
   Because Dict[str, Any] is not supported by most languages, because of lower safety.

2. sometimes it is effective to declare typing variables as a global variable to comprehend inputs as a same group.

For python3.8 >=,

3. Generic class
   We can use something like template in C++
   We can separate Union depending on files.

4. Literal
   we can accept only specific variable rather than typing

5. Final
   If we do not hope the change in variables, we can use it.

6. TypedDict
   We can define meta class for dict.

Also we need to discuss (with Frank as well) when we should stop the support of older python versions.
Just in case, python3.6 is already not supported officially and python3.7 will be not supported from this December.

[codecov]

Codecov Report

Merging #240 (64ce7e5) into development (1818445) will increase coverage by 0.79%.
The diff coverage is 83.02%.

@@               Coverage Diff               @@
##           development     #240      +/-   ##
===============================================
+ Coverage        80.82%   81.62%   +0.79%     
===============================================
  Files              148      150       +2     
  Lines             8563     8625      +62     
  Branches          1323     1325       +2     
===============================================
+ Hits              6921     7040     +119     
+ Misses            1163     1108      -55     
+ Partials           479      477       -2     

Impacted Files	Coverage Δ
autoPyTorch/api/tabular_regression.py	96.87% <ø> (ø)
autoPyTorch/data/base_target_validator.py	97.95% <ø> (-0.05%)	⬇️
autoPyTorch/data/tabular_feature_validator.py	86.30% <ø> (-0.09%)	⬇️
...ts/setup/network_backbone/base_network_backbone.py	88.63% <0.00%> (ø)
...components/setup/network_head/base_network_head.py	96.15% <0.00%> (ø)
...Torch/pipeline/components/training/metrics/base.py	65.67% <ø> (ø)
...PyTorch/pipeline/traditional_tabular_regression.py	25.86% <25.86%> (ø)
...ine/components/setup/network_embedding/__init__.py	64.89% <44.44%> (+0.76%)	⬆️
autoPyTorch/evaluation/abstract_evaluator.py	75.89% <46.87%> (-2.28%)	⬇️
...h/pipeline/components/training/trainer/__init__.py	70.81% <60.00%> (+0.48%)	⬆️
... and 102 more

---

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 eda89f4...64ce7e5. Read the comment docs.

[franchuterivera]

Thanks for the PR. I left some questions. Let me know what you think about them.

[nabenabe0928]

You can declare a temporal variable for the types.
Anyways, the variable scope here is small enough and it will increase the readability.

[ravinkohli]

So, we already have a variable for the types called SUPPORTED_TARGET_TYPE which we can use directly using isinstance checks, however, to check for sparsity we do not use isinstance rather we use scipy.issparse and hence I did not think it was appropriate to make another variable just for this type check.