feat(agentcore): add hindsight-agentcore integration for Bedrock AgentCore Runtime

[benfrank241]

Summary

Adds hindsight-agentcore — a Python package providing durable cross-session memory for Amazon Bedrock AgentCore Runtime agents.

AgentCore Runtime sessions are explicitly ephemeral (terminate on inactivity, reprovision fresh). This adapter makes Hindsight the durable memory layer that survives session churn.

Key design decisions from the integration spec:

• Memory is keyed to stable user identity, never runtimeSessionId — sessions expire, banks don't
• Default bank format: tenant:{tenant_id}:user:{user_id}:agent:{agent_name}
• Retention is async by default — never delays the turn response
• Fails closed on bank resolution errors — no cross-user memory leakage
• Graceful degradation everywhere — memory failures never surface to the user turn

Package structure

hindsight-integrations/agentcore/
├── hindsight_agentcore/
│   ├── __init__.py       # Public API
│   ├── adapter.py        # HindsightRuntimeAdapter (before_turn/after_turn/run_turn)
│   ├── bank.py           # TurnContext + bank resolution
│   ├── config.py         # configure() / get_config() / reset_config()
│   └── errors.py         # HindsightAgentCoreError, BankResolutionError
└── tests/
    ├── test_adapter.py   # 19 tests
    ├── test_bank.py      # 10 tests
    └── test_config.py    # 12 tests

API surface

from hindsight_agentcore import HindsightRuntimeAdapter, TurnContext, configure

configure(hindsight_api_url="...", api_key="...")

adapter = HindsightRuntimeAdapter(agent_name="support-agent")

# High-level: recall → execute → retain
result = await adapter.run_turn(context, payload, agent_callable=my_agent)

# Low-level hooks
memory_context = await adapter.before_turn(context, query=user_message)
result = await my_agent(payload, memory_context)
await adapter.after_turn(context, result=result["output"], query=user_message)

Test plan

• uv run pytest — 41 tests pass
• ruff check — clean
• Integration smoke test against live Hindsight instance
• End-to-end test with an actual AgentCore Runtime invocation
• Docs page (follow-up PR)