feat: add langchain deepagents backend integration

Adds langchain-agent-sandbox package that implements the DeepAgents
BackendProtocol by wrapping a SandboxClient.

Features:
- Execute commands in sandbox
- Read/write/edit files with path virtualization under /app
- Grep and glob operations
- File upload/download
- Async variants of all operations
- Path traversal protection

The package is self-contained in clients/python/langchain-agent-sandbox/
with its own tests and mypy configuration.

Author: johannhartmann

clients/python/langchain-agent-sandbox/README.md

@@ -0,0 +1,29 @@
+# LangChain Agent Sandbox Backend
+
+This package provides a LangChain DeepAgents backend that connects to the Kubernetes agent-sandbox runtime via the existing Python SDK.
+
+## Install
+
+```sh
+pip install langchain-agent-sandbox
+```
+
+## Usage
+
+```python
+from agentic_sandbox import SandboxClient
+from deepagents import create_deep_agent
+from langchain_agent_sandbox import AgentSandboxBackend
+
+with SandboxClient(template_name="my-template", namespace="default") as client:
+    backend = AgentSandboxBackend(client, root_dir="/app")
+    agent = create_deep_agent(backend=backend)
+
+    result = agent.invoke("List files in the sandbox")
+    print(result)
+```
+
+## Notes
+- The backend wraps the `SandboxClient` methods (`run`, `read`, `write`) to implement the DeepAgents protocol.
+- File paths are virtualized under `root_dir` (default `/app`).
+- Requires POSIX utilities in the sandbox image: `sh`, `base64`, `grep`, `find`, `mkdir`, `ls`, and `test`.

clients/python/langchain-agent-sandbox/langchain_agent_sandbox/__init__.py

@@ -0,0 +1,3 @@
+from .backend import AgentSandboxBackend
+
+__all__ = ["AgentSandboxBackend"]