Add `SkillExecutor.External()` skill script execution mode

[SergeyMenshykh]

Summary

Add support for executing agent skill scripts via a user-provided executor, encapsulated as SkillExecutor.External().

Background

With the SkillExecutor abstract class and HostedCodeInterpreter mode established in #4131, this issue adds the second execution mode where script execution is delegated to a user/developer-provided callback. This is the second of three planned execution modes:

1. SkillExecutor.HostedCodeInterpreter() — LLM code interpreter (Add SkillExecutor.HostedCodeInterpreter() skill script execution mode #4131)
2. SkillExecutor.External() ← this issue
3. SkillExecutor.Hybrid() — conditional routing between the two (future)

Requirements

• Add a static External() factory method on SkillExecutor that accepts a user-provided execution delegate/predicate and returns a concrete sealed subclass instance.
• The External mode (concrete sealed subclass) must:
  • Provide instructions that direct the LLM to call the execute_script tool for script execution.
  • Register an execute_script tool as a provided tool.
  • When execute_script is invoked, delegate execution to the user-provided callback, passing relevant context (script content, skill name, etc.).
• The user-provided delegate should receive sufficient context to execute the script and return a result.
• Add a sample project Agent_Step03_ScriptExecutionWithExternalExecutor demonstrating usage of this execution mode.
• The execute_script tool should be wrapped with ApprovalRequiredAIFunction by default to require user approval before executing scripts externally.
  • The External() factory method should accept an optional parameter to disable approval (e.g., requireApproval: bool = true). When enabled: new
    ApprovalRequiredAIFunction(AIFunctionFactory.Create(executeScript)). When disabled, the tool is registered directly.

Acceptance Criteria

• External() static factory method on SkillExecutor accepts a user-provided execution delegate
• External mode is a sealed concrete subclass of SkillExecutor
• External mode provides instructions directing the LLM to use execute_script
• External mode registers an execute_script tool
• execute_script tool invocations delegate to the user-provided callback
• Unit tests cover mode initialization, instruction/tool provisioning, and delegate invocation
• Integration with FileAgentSkillsProvider works consistently with the HostedCodeInterpreter mode
• Agent_Step03_ScriptExecutionWithExternalExecutor sample project added
• execute_script tool is wrapped with ApprovalRequiredAIFunction by default
  • Approval can be disabled via External(execute, requireApproval: false)
  • Unit tests cover both approval-enabled and approval-disabled paths

Part of #3499
Depends on #4131