Add the ability to run agent sessions interactively within the sandbox container, allowing users to iterate on tasks with direct terminal access to the AI agent.

Closes #310

Changes

• Added --interactive (-i) flag to the iterate command to start an interactive session instead of running in background
• Implemented runInteractive() method in Docker and Podman sandbox classes to launch containers with stdio: 'inherit'
• Created new session command in rover-agent package that runs agents in interactive mode with pre-context support
• Refactored the entrypoint script into a configurable template that supports both autonomous and interactive modes
• Consolidated entrypoint generation by adding includeTaskSetup parameter to generateEntrypoint() method
• Added validation to prevent iterating over running tasks
• Improved signal handling in entrypoint script to clean up temporary context files on exit

Notes

The interactive mode creates a temporary .rover-context directory in the workspace to share context files with the agent. This directory is automatically cleaned up when the session ends.

The entrypoint script now uses template variables ({validateTaskFileFunction}, {taskDataSection}, etc.) to conditionally include task-related setup based on the execution mode.