[Feature] Allow multipe inputs to models

[matteobettini]

This PR allows multiple inputs to models (given from their input spec)

The current models will behave in the following way:

• MLP will check that all inputs have the same shape up to the last dim and cat them before feeding them to the NN
• GNN will to the same
• CNN will accept 2 types of inputs: images and 1-dim tensors. It will cat the images and process them and then cat them to the tensors before the MLP layer

In this PR we move the concatenation of actions and state (that was happening in the SAC and DDPG algorithms) to the model by directly feeding both inputs

cc @ezhang7423

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 87.50000% with 25 lines in your changes are missing coverage. Please review.

Project coverage is 53.83%. Comparing base (e6afcf5) to head (f4118c5).

Files	Patch %	Lines
benchmarl/models/cnn.py	88.88%	13 Missing :warning:
benchmarl/models/common.py	60.00%	4 Missing :warning:
benchmarl/models/mlp.py	76.92%	3 Missing :warning:
benchmarl/models/gnn.py	81.81%	2 Missing :warning:
benchmarl/algorithms/maddpg.py	0.00%	1 Missing :warning:
benchmarl/algorithms/masac.py	0.00%	1 Missing :warning:
test/test_models.py	97.72%	1 Missing :warning:

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #73 +/- ## =========================================== - Coverage 89.99% 53.83% -36.17% =========================================== Files 69 70 +1 Lines 2739 2907 +168 =========================================== - Hits 2465 1565 -900 - Misses 274 1342 +1068 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.