feat(weixin): add split_messages config to control message splitting …

[guantoubaozi]

Title:

feat(weixin): add split_messages config to control message splitting behavior

Description:

## Problem

The Weixin adapter splits every response into multiple messages — one per paragraph or top-level line break. While this was designed for readability, in practice it makes responses appear as a flood of separate message bubbles in WeChat, which many users find noisy and hard to follow.

Currently there is no way to control this behavior.

## Solution

Add a new `split_messages` configuration option for the Weixin adapter:

| Value | Behavior |
|-------|----------|
| `true` (default) | Split at paragraph boundaries — **current behavior, unchanged** |
| `false` | Keep responses in a single message when under `MAX_MESSAGE_LENGTH` (4000 chars) |

Oversized content (> 4000 chars) always falls back to block-aware splitting regardless of this setting.

### Configuration

Via `config.yaml`:

```yaml
platforms:
  weixin:
    extra:
      split_messages: false
```

Or via environment variable:

```env
WEIXIN_SPLIT_MESSAGES=false
```

## Changes

- `WeixinAdapter.__init__()`: Read `split_messages` from `extra` config or `WEIXIN_SPLIT_MESSAGES` env var (default: `true`)
- `WeixinAdapter._split_text()`: When `split_messages` is `false`, skip per-line splitting and return content as a single message (or use block-aware packing for oversized content)

## Backward Compatibility

Default is `true`, so existing behavior is completely unchanged. Users who prefer merged messages can opt in by setting the config.

## Testing

Tested with personal WeChat (iLink Bot API) in production gateway:

- `split_messages: true` (default): responses split at paragraph boundaries ✅
- `split_messages: false`: short responses delivered as single message ✅
- `split_messages: false` + long response (> 4000 chars): split at block boundaries ✅
- Code fences: preserved intact in both modes ✅
- Markdown formatting: renders correctly in both modes ✅

[teknium1]

Merged via salvage PR #7903. Your contribution — simplest core fix — was incorporated into the combined fix. All six PRs addressing this issue were synthesized into a single implementation that takes the best aspects of each. Thank you @guantoubaozi for the contribution!