[Feature]: cron定时任务建议可以选择新开subagent并选择其他模型

## Summary

It is recommended that the cron scheduled task feature support selecting to create a new subagent and specifying other non-default models for task execution, to meet the differentiated model capability requirements of different scheduled tasks.

## Component(s) Affected

- [x] Core / Backend (app, agents, config, providers, utils, local_models)

## Problem / Motivation

Currently, cron scheduled tasks use the main agent and a fixed model for execution by default. It is impossible to adapt different model capabilities for scheduled tasks in different scenarios (e.g., data analysis tasks require more powerful reasoning models, while simple notification tasks can use lightweight models), nor can independent subagents be used to isolate the execution context of scheduled tasks. This optimization will allow users to more flexibly configure the execution carrier of scheduled tasks, improve the adaptability and isolation of task execution, and all users who use cron scheduled tasks will benefit from it.

## Proposed Solution

1. Add two optional configurations in the configuration interface/items of cron scheduled tasks:
- "Execution Agent": Support selecting "Main Agent" or "Create New Subagent"; when selecting to create a new subagent, users can customize the subagent name;
- "Execution Model": Support drop-down selection of all available models configured in the system (including local models and third-party integrated models), and the default model of the current main agent is used by default.
2. When triggering scheduled tasks, the backend creates/calls the corresponding subagent according to the configured agent type, and specifies the selected model to execute the task;
3. The newly added configuration items are non-mandatory, and the original default logic is retained to ensure that the execution of existing scheduled tasks is not affected.

## Alternatives Considered

1. Let users manually create a subagent and then fill in the subagent ID in the scheduled task for association: This method has a high operation threshold, is not intuitive, and is unfriendly to non-technical users;
2. Only support switching models without supporting the creation of new subagents: It cannot solve the problem of isolating the task execution context, and the execution records and contexts of different scheduled tasks will interfere with each other.