feat: migrate pybaum to optree

[abelaba]

Summary

• Replace pybaum with optree for all pytree operations used in optimagic (tree_just_flatten,tree_unflatten,tree_flatten,tree_map, tree_equal, leaf_names)
• Remove pybaum as a dependency and add optree>=0.19.0,<0.20.
• tree_registry.py: Rewritten to use optree's register_pytree_node with a custom namespace, including support for pandas DataFrame/Series, and JAX arrays.
• typing.py: Added shared namespace constants (optree_namespaces list and default_namespace).

Why wrapper methods instead of using optree directly

1. Flattening: optree does not preserve insertion order of dictionaries by default. Without the wrapper, we would need to use first convert dictionaries to an OrderedDict (referenced in this doc). The wrapper ensures dict
   ordering is maintained. (see failing test screenshots below if order is not maintained for dictionary)

In addition, ordering currently works for flattening dictionaries when a namespace is provided.

2. Unflattening: pybaum and optree have different Tree structures. The tree_unflatten wrapper maintains backward compatibility for methods that use pybaum's tree structure.

Remaining

• Adding type hints and docstrings for newly added methods.
• Flattening and unflattening for torch tensors.

Test plan

• All existing fast tests pass with the new optree backend.