add mnle as model class to snle

[janfb]

this provides easy access to Mixed Neural Likelihood Estimation (MNLE) in sbi: MNLE is a child of SNLE and can be used by calling MNLE() and otherwise using the same API as with SNLE:

from sbi.inference import MNLE
trainer = MNLE(prior)
estimator = trainer.append_simulations(theta, x).train()
posterior = trainer.build_posterior()

It is implemented as a single density estimator class that has two separate nets for discrete and continuous data, but trains them with a single loss call, i.e., using the usual SNLE train method (thanks to @michaeldeistler for the idea).

It requires a separate mixed_likelihood_estimator_based_potential because it implements its own log_prob for iid-x to get some speed up during MCMC.

Tasks:

• finish docstrings and comments
• add interface to potential function for efficient evaluation with iid-trial data
• add tutorial notebook for inference in decision-making models.

[codecov-commenter]

Codecov Report

Merging #638 (8e464b3) into main (78cc11e) will increase coverage by 1.71%.
The diff coverage is 89.34%.

@@            Coverage Diff             @@
##             main     #638      +/-   ##
==========================================
+ Coverage   74.81%   76.53%   +1.71%     
==========================================
  Files          75       77       +2     
  Lines        5667     5826     +159     
==========================================
+ Hits         4240     4459     +219     
+ Misses       1427     1367      -60     

Flag	Coverage Δ
unittests	76.53% <89.34%> (+1.71%)	⬆️

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

Impacted Files	Coverage Δ
sbi/inference/potentials/__init__.py	100.00% <ø> (ø)
sbi/inference/snle/snle_a.py	100.00% <ø> (ø)
sbi/utils/user_input_checks_utils.py	90.25% <80.00%> (-1.80%)	⬇️
sbi/inference/snle/mnle.py	85.71% <85.71%> (ø)
sbi/neural_nets/mnle.py	88.11% <88.11%> (ø)
sbi/inference/__init__.py	100.00% <100.00%> (ø)
sbi/inference/posteriors/mcmc_posterior.py	74.82% <100.00%> (+1.58%)	⬆️
...inference/potentials/likelihood_based_potential.py	100.00% <100.00%> (ø)
sbi/inference/snle/__init__.py	100.00% <100.00%> (ø)
sbi/inference/snle/snle_base.py	91.83% <100.00%> (-0.17%)	⬇️
... and 12 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 78cc11e...8e464b3. Read the comment docs.

[michaeldeistler]

I like the implementation a lot, thanks! I left some comments below

[michaeldeistler]

Thanks a lot! Only minor comments below. Good to go once they are addressed.