Add stop sequences and prefill options to Claude 3 models

[irthomasthomas]

• Introduced stop_sequences option in ClaudeOptions to allow users to specify sequences that will terminate text generation.
• Implemented prefill option to provide a starting point for text generation.
• Updated README.md to document new features with usage examples for stop sequences and prefill.
• Added JSON validation in ClaudeOptions for stop_sequences to ensure proper format.
• Modified message construction in ClaudeMessages class to include prefill text when provided.

Files changed:

• README.md
• llm_claude_3.py

[irthomasthomas]

Notes:

Neither the prefill, not the stop sequence are included in the response.

llm -m claude-3.5-sonnet -o prefill "<PLAN>" "Surprise me with  a dbus shell function" -o stop_sequences '["</PLAN>"]'

1. Create a D-Bus shell function that does something unexpected or interesting
2. Include comments explaining what the function does
3. Make sure it's a valid shell function that can be used in a bash script
4. Add some error handling and parameter checking

{"stop_sequences": ["</PLAN>"], "prefill": "<PLAN>", "max_tokens": 8192, "temperature": 1.0}

If you use multiple stop_sequences, you can determine which was received from the response_json:

{
    "id": "msg_01MTH7xDebox1v8VPqH9nVoG",
    "content": [
        {
            "text": "\n1. Create a D-Bus shell function that does something unexpected or interesting\n2. Include comments explaining what the function does\n3. Make sure it's a valid shell function that can be used in a bash script\n4. Add some error handling and parameter checking\n",
            "type": "text"
        }
    ],
    "model": "claude-3-5-sonnet-20240620",
    "role": "assistant",
    "stop_reason": "stop_sequence",
    "stop_sequence": "</PLAN>",
    "type": "message",
    "usage": {
        "input_tokens": 21,
        "output_tokens": 60
    }
}