[Bug] Feishu channel: filter_thinking and filter_tool_messages config not working

## Bug Description

Even though `filter_thinking: true` and `filter_tool_messages: true` are configured in `config.json` for the Feishu (Lark) channel, the Agent still outputs thinking processes and tool call details to the Feishu chat session.

## Current Configuration

```json
{
"feishu": {
"enabled": true,
"filter_tool_messages": true,
"filter_thinking": true
}
}
```

Verified with `copaw channels list`:

```
Feishu [enabled]
filter_tool_messages: True
filter_thinking : True
```

## Expected Behavior

- `filter_thinking: true` → Should NOT output thinking processes
- `filter_tool_messages: true` → Should NOT output tool call details

## Actual Behavior

Even with the above configuration, the Feishu session still shows:
- Detailed tool call outputs
- Thinking process/explanation text

## Reproduction Steps

1. Configure `filter_tool_messages: true` and `filter_thinking: true` for the Feishu channel
2. Execute a task that requires tool calls (e.g., uploading whiteboard diagrams)
3. Observe the Feishu session output

**Result**: Tool call details and thinking processes are still visible

## Environment

- **CoPaw Version**: 1.0.1
- **Channel**: Feishu (Lark)
- **OS**: Linux 6.14.0-37-generic (x86_64)

## Related Issues/PRs

- **#2972** (Open): [Feature Request] 面板聊天记录显示思考过程，无法关闭 - https://github.com/agentscope-ai/CoPaw/issues/2972
- **#2756** (Closed, merged today): fix(console): preserve filter_thinking and filter_tool_messages when saving console channel - https://github.com/agentscope-ai/CoPaw/pull/2756

## Root Cause Analysis

PR #2756 fixed a similar issue for the Console channel today (2026-04-07), where saving channel configuration would incorrectly override the filter settings. The Feishu channel likely has the same configuration save/apply logic bug.

## Request

Please investigate and apply a similar fix to the Feishu channel as was done for the Console channel in PR #2756.

## Temporary Workaround

Currently, users must tolerate the verbose output or manually ignore the tool call information. The functionality is not affected, only the output verbosity.