Added version ceiling for anthropic token count.

[3coins]

Description

Current get_num_tokens method in ChatBedrock utilizes a legacy Anthropic API count_tokens, support for which has been dropped by Anthropic in their latest release v0.39.0. This PR adds a version ceiling for the anthropic client, only supporting token count for anthropic client versions 0.38.0 and lower. Users who have a need to use token counting in their applications should follow these steps.

1. Stop using the get_num_tokens from the ChatBedrock class.
2. Start using get_num_tokens_from_messages from the ChatAnthropic class.

Here is some sample code that follows this strategy.

from langchain_core.messages import HumanMessage
from langchain_aws import ChatBedrock
from langchain_anthropic import ChatAnthropic

def main():
    messages = [
        HumanMessage("What is the capital of France?")
    ]

    llm = ChatBedrock(
        model="anthropic.claude-3-5-sonnet-20241022-v2:0"
    )

    token_counter_model = ChatAnthropic(
        model="claude-3-5-sonnet-20241022"
    )

    token_count = token_counter_model.get_num_tokens_from_messages(messages)
    print(f"token count is {token_count}")

    if token_count < 100:
        print(llm.invoke(messages))
    else:
        print(
            "Warning: Trim to smaller prompt to reduce token count."
        )

if __name__ == "__main__":
    main()

References

Anthropic Documentation for count_tokens API
https://docs.anthropic.com/en/api/messages-count-tokens

Dropped support for count_tokens in AnthropicLLM
https://github.com/langchain-ai/langchain/blob/master/libs/partners/anthropic/langchain_anthropic/llms.py#L375

New implementation in ChatAnthropic
https://github.com/langchain-ai/langchain/blob/master/libs/partners/anthropic/langchain_anthropic/chat_models.py#L1117