Add conditional optimal transport for flow matching

[arrjon]

This pull request introduces significant improvements to the optimal transport functionality, with a focus on supporting partial and conditional optimal transport in both the Sinkhorn and Log-Sinkhorn algorithms.

• Added support for partial optimal transport and conditional optimal transport in both Sinkhorn and Log-Sinkhorn algorithms. Both can in principle learn a better transport map for conditional flow matching.
• Refactored the Sinkhorn and Log-Sinkhorn methods to improve computational efficiency (using the dual optimization problem).
• Updated the optimal_transport API to accept and correctly propagate conditions, and to return resampled conditions and assignments for both balanced and unbalanced (partial/conditional) settings.
• Changed default configuration values and documentation to reflect new options and recommended settings for conditional optimal transport.
• Refactored code in flow_matching.py to handle new outputs from the optimal transport functions and to properly resample conditions when provided.

[codecov[bot]]

Codecov Report

:x: Patch coverage is 97.00599% with 5 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
bayesflow/utils/optimal_transport/log_sinkhorn.py	95.34%	2 Missing :warning:
bayesflow/utils/optimal_transport/sinkhorn.py	95.65%	2 Missing :warning:
bayesflow/networks/flow_matching/flow_matching.py	0.00%	1 Missing :warning:

Files with missing lines	Coverage Δ
...sflow/utils/optimal_transport/optimal_transport.py	100.00% <100.00%> (ø)
bayesflow/utils/optimal_transport/ot_utils.py	100.00% <100.00%> (ø)
bayesflow/networks/flow_matching/flow_matching.py	94.35% <0.00%> (+1.49%)	:arrow_up:
bayesflow/utils/optimal_transport/log_sinkhorn.py	96.61% <95.34%> (-3.39%)	:arrow_down:
bayesflow/utils/optimal_transport/sinkhorn.py	96.87% <95.65%> (-3.13%)	:arrow_down: