Support custom sandboxes

**Is your feature request related to a problem? Please describe.**
I'd like to be able to easily plug in a custom sandbox. However, the e2b and local python executor are currently hardcoded.

**Describe the solution you'd like**
I'd like to have a reusable abstraction for what a sandbox should support, so I could implement a custom sandbox and pass it as a parameter to the agent.

**Is this not possible with the current options.**
Currently the available sandboxes are hardcoded.

**Describe alternatives you've considered**
It's possible to hack together running the whole agent inside a custom sandbox, however this solution ties between the execution environment of the agent itself and the environment in which the tools run. It's a problem for example when it's needed to have a specific set of dependencies for a run.

The issue names no files or tests; start by locating where the e2b and local Python executors are hardcoded and how the agent receives execution options. Define the sandbox capabilities and parameter boundary, then verify that a custom sandbox can be supplied without coupling the agent's environment to the tool execution environment.

Written by the indexing model from the issue text.