This commit introduces the `resize_token_embeddings` function for the
`OLMoForCausalLM` class. The function is based on the implementation
from the Hugging Face Transformers library, with modifications to
accommodate the specific requirements of the OLMo model.

The `resize_token_embeddings` function allows resizing the input token
embeddings matrix of the model when the number of tokens differs from
the model's `embedding_size` configuration. It updates the `embedding_size`
attribute in both the model configuration and the model itself, ensuring
consistency after resizing.

The function also handles tying the weights of the input and output
embeddings if the model supports weight tying.

Attribution:
The implementation of this function is inspired by and adapted from
the `resize_token_embeddings` function in the Hugging Face Transformers
library (https://github.com/huggingface/transformers). The original code
is licensed under the Apache License 2.0.

Modifications:
- Updated the function to use the `embedding_size` attribute instead of
  `vocab_size` to align with the OLMo model configuration.
- Adjusted the docstring and comments to match the OLMo model's terminology
  and requirements.

clarifies difference between the updated method and the base class method

switching warning to log.warning and fixes comment indent