[Fix] register_checkpoint_saver API was not stable under TF version 2.11.

[MoFHeka]

Description

[Fix] register_checkpoint_saver API was not stable under TF version 2.11. Therefore, for compatibility considerations, before version 2.14, SingleDeviceSaver was used to save.

Here is the error looks like when using TF 2.9

AssertionError: Tried to export a function which references 'untracked' resource Tensor("movie_id_DELayer-parameter/DynamicEmbedding/movie_id_DELayer-shadow/m/cond/movie_id_DELayer-parameter_DynamicEmbedding_movie_id_DELayer-shadow_m_mht_1of1:0", shape=(), dtype=resource). TensorFlow objects (e.g. tf.Variable) captured by functions must be 'tracked' by assigning them to an attribute of a tracked object or assigned to an attribute of the main object directly. See the information below:
        Function name = b'__inference_restore_fn_5544'
        Captured Tensor = Tensor("movie_id_DELayer-parameter_DynamicEmbedding_movie_id_DELayer-shadow_m_mht_1of1:0", shape=(), dtype=resource, device=/job:localhost/replica:0/task:0/device:CPU:0)
        Trackable referencing this tensor = <tensorflow_recommenders_addons.dynamic_embedding.python.ops.cuckoo_hashtable_ops.CuckooHashTable object at 0x7f4630380a30>

Type of change

• Bug fix
• New Tutorial
• Updated or additional documentation
• Additional Testing
• New Feature

Checklist:

• I've properly formatted my code according to the guidelines
  • By running yapf
  • By running clang-format
• This PR addresses an already submitted issue for TensorFlow Recommenders-Addons
• I have made corresponding changes to the documentation
• I have added tests that prove my fix is effective or that my feature works

How Has This Been Tested?

This bug causes the code of the latest main branch to fail to save any DE variable and raise error under TF2.9:

model.fit()
model.save(export_dir, overwrite=True, options=saved_options)