Feature: allow use of Keras functional API by sanitizing build input shapes

[vpratz]

Fixes #329.

This PR adds a decorator sanitize_input_shape to replace the first entry of input_shape with a dummy batch size of 32 in case it is None.
The decorator is added to the relevant network build functions.

If you have any ideas/wishes regarding naming, please let me know and I'll adjust it.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

Attention: Patch coverage is 86.36364% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
bayesflow/utils/decorators.py	70.00%	3 Missing ⚠️

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Files with missing lines	Coverage Δ
bayesflow/networks/deep_set/deep_set.py	100.00% <100.00%> (ø)
bayesflow/networks/deep_set/equivariant_module.py	97.22% <100.00%> (+0.16%)	⬆️
bayesflow/networks/deep_set/invariant_module.py	94.44% <100.00%> (+0.32%)	⬆️
bayesflow/networks/lstnet/lstnet.py	100.00% <100.00%> (ø)
bayesflow/networks/lstnet/skip_recurrent.py	100.00% <100.00%> (ø)
bayesflow/networks/summary_network.py	51.85% <100.00%> (+3.85%)	⬆️
bayesflow/utils/decorators.py	74.32% <70.00%> (+0.88%)	⬆️

[LarsKue]

Looks good to me. Ideally, in the long term, we should move away from the (expensive and unreliable) way of using self.call in self.build, though.

[vpratz]

Thanks for the quick review!