feat(security): add max_context_messages to prevent retry amplificati…

[Waqar53]

…on (#2056)

This PR addresses the retry amplification security issue described in #2056.

Problem

During retries, each failed attempt appends 2 messages (assistant response + tool error). With 10 retries, this can cause 506x context growth, leading to:

• Token budget exhaustion
• Cost amplification
• Potential denial of service

Solution

Added parameter that limits the number of messages kept in context during retries. The system message is always preserved.

Usage

client = instructor.from_openai(OpenAI())

response = client.chat.completions.create(
    model='gpt-4',
    response_model=MyModel,
    max_retries=10,
    max_context_messages=20,  # Limit context to 20 messages
    messages=[...]
)

Changes

• Added truncate_context_messages() function in instructor/core/retry.py
• Added max_context_messages parameter to retry_sync() and retry_async()
• Exposed parameter through patch.py for user access
• Added comprehensive unit tests (10 test cases)

Testing

• All tests pass
• Backwards compatible (default is None = no truncation)