Different built-in sandbox types and their distinctions

**Description**

I'm currently evaluating AgentScope Runtime for our project and trying to understand the differences between the various built-in sandbox types. I've read through the documentation but still have some questions about when to use each type and what specific capabilities they provide.

**Background**

From the documentation, I understand that AgentScope Runtime provides multiple built-in sandbox types, including `base`, `gui`, `browser`, `filesystem`, and `mobile`. Each uses a different Docker image:

- `runtime-sandbox-base:latest`
- `runtime-sandbox-gui:latest`
- `runtime-sandbox-browser:latest`
- `runtime-sandbox-filesystem:latest`
- `runtime-sandbox-mobile:latest`

**Questions**

1. **Capabilities**: Could you provide a detailed comparison of what each sandbox type can and cannot do? For example:
- Does `BaseSandbox` only support Python code and shell commands?
- Which types support VNC/ graphical interfaces?
- What are the specific tool methods exposed by each type?

2. **Image Relationships**: Are the specialized sandbox images (e.g., `browser`, `filesystem`) built on top of the `base` image, or are they completely independent? If I pull the `browser` image, do I still need the `base` image?

3. **Resource Consumption**: What are the typical resource requirements (CPU, memory, disk) for each sandbox type? The `mobile` sandbox uses Docker-in-Docker (DinD) to run an Android emulator【0†L?】— does it require significantly more resources?

Thank you for your time and for building this great project!