Fix failing tutorials, change MNLE default for log_transform to False

[janfb]

The change to detecting NaNs in all losses already payed off: In MNLE, we were using the log_transform=True per default. This was performing better and makes sense for positive data like reaction times in decision-making experiments. However, it does not make sense in general. In implemented methods tutorial we were passing Gaussian data, which then caused NaNs after the log-transform, causing the loss to become NaN.

This PR

• changes the default to False and add small improvements to the tutorial notebook.
• adapt all tutorial and example notebooks to explicitly apply the log transform when dealing with bounded decision-making data
• fix bug in Decision-Making Example notebook that caused the time-our error in Continuous-Deployment (main tests failing)

@manuelgloeckler / @michaeldeistler can you have a look please?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.23%. Comparing base (d3f22b5) to head (40a7dde).
Report is 13 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1367      +/-   ##
==========================================
- Coverage   89.40%   89.23%   -0.18%     
==========================================
  Files         118      118              
  Lines        8715     8779      +64     
==========================================
+ Hits         7792     7834      +42     
- Misses        923      945      +22     

Flag	Coverage Δ
unittests	89.23% <ø> (-0.18%)	⬇️

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

Files with missing lines	Coverage Δ
sbi/neural_nets/net_builders/mnle.py	100.00% <ø> (ø)

... and 6 files with indirect coverage changes

[michaeldeistler]

LGTM!