[BUG] UnboundLocalError in AEAttentionBiGRUNetwork When temporal_latent_space=True

[lucifer4073]

Describe the bug

When initializing AEAttentionBiGRUNetwork with temporal_latent_space=True, calling build_network results in an UnboundLocalError. The issue occurs because the variable shape_before_flatten is referenced but never assigned a value when temporal_latent_space=True. The issue does not occur when temporal_latent_space=False.

Steps/Code to reproduce the bug

from aeon.networks import AEAttentionBiGRUNetwork  

# Initialize with temporal_latent_space=True
aeattentionbigru = AEAttentionBiGRUNetwork(latent_space_dim=64, temporal_latent_space=True)  
aeattentionbigru.build_network((1000, 5))  # This causes the error

Expected results

The build_network function should execute without errors when temporal_latent_space=True, correctly constructing the network.

Actual results

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
line 177, in build_network
    shape=shape_before_flatten, name="decoder_input"
          ^^^^^^^^^^^^^^^^^^^^
UnboundLocalError: cannot access local variable 'shape_before_flatten' where it is not associated with a value

Versions

No response

[4444Harsh]

The issue was reported and addressed in the following pull requests:

[ENH] Test coverage for AEAttentionBiGRUNetwork Improved #2540

[BUG] Bug resolved in Aeabgru for temporal_latent_space=True #2566
The bug is a result of an unassigned variable in a specific code path. Ensuring that shape_before_flatten is assigned a value in all cases will resolve the UnboundLocalError and allow the network to be built correctly when temporal_latent_space=True.