Agent-tool integration

LMTask specializes and serves a model-backed task. An orchestration framework can then register that task as one of its tools.

The example remains framework-neutral so LMTask does not require LangChain, LangGraph, or another agent runtime:

from sentiment_tool import classify_sentiment

result = classify_sentiment(
    'The documentation was clear, but model startup was slow.')
print(result)

The callable can be registered with any framework that accepts Python functions or JSON-compatible tool results. Keep planning, state, and tool selection in the orchestration layer; keep model loading, task prompts, response parsing, caching, and specialization in LMTask.