feat(langchain): add HealthMiddleware for agent health monitoring and resilience

[ixchio]

Add built-in health-aware middleware for LangChain v1 agents that monitors tool/model health and provides automatic retry, fallback, and telemetry capabilities.

Fixes #34193

Features

• HealthStats: Rolling window statistics tracking per-tool/model (error rate, P95 latency)
• HealthPolicy: Configurable degradation thresholds
• HealthMiddleware: Auto retry with exponential backoff, model fallback, tool disabling
• Telemetry: Pluggable emitter with create_langsmith_emitter() helper
• Persistence: Optional store integration for cross-run stats via store param
• Provider tool preservation: Keep provider dict tools active even when degraded

Usage

from langchain.agents.middleware import HealthMiddleware, HealthPolicy

health = HealthMiddleware(
    policy=HealthPolicy(max_error_rate=0.2, consecutive_failures=3),
    fallback_model="openai:gpt-4o-mini",
    disable_tools=True,
)

agent = create_agent(model="openai:gpt-4o", tools=[...], middleware=[health])

Testing
35 unit tests covering HealthStats, HealthPolicy, HealthMiddleware
All 534 middleware tests pass

[sydney-runkle]

Thanks for your contribution!

At this time, we're only adding middlewares that we've seen significant demand for and have the capacity to maintain.

Cool implementation though, we're definitely seeing lots of middlewares that add resilience to agentic workflows!