Document LSTMCell carry as the (c, h) tuple it actually is

[adityasingh2400]

Summary

The LSTMCell.__call__ docstring described carry only as "the hidden state of the LSTM cell", which is misleading: carry is actually a tuple (c, h) of the cell state and the hidden state, as initialize_carry and the implementation both make clear (return (c, h) and c, h = carry).

This PR spells out the contract in the docstring:

carry: a tuple (c, h) of the cell state c and the hidden state h, both of shape (*batch, features). Typically initialized using LSTMCell.initialize_carry.

Applied to all four affected __call__ docstrings for parity:

• flax/linen/recurrent.py LSTMCell
• flax/linen/recurrent.py OptimizedLSTMCell
• flax/nnx/nn/recurrent.py LSTMCell
• flax/nnx/nn/recurrent.py OptimizedLSTMCell

The OptimizedLSTMCell docstrings previously pointed at LSTMCell.initialize_carry; those references are also updated to OptimizedLSTMCell.initialize_carry to match the class users actually have a handle to.

Fixes #4124.

Test plan

• Pure docstring change, no runtime behavior modified.
• No public API changes.
• Wording matches the implementation (c, h = carry, return (new_c, new_h), new_h) and initialize_carry shapes.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.