fix(weixin): prevent excessive message splitting on WeChat platform

[sherunlock03]

Problem

The original implementation of _split_text_for_weixin_delivery() in the WeChat (Weixin) platform adapter was splitting messages at every newline character. This caused messages that could have been delivered as a single unit to be fragmented into multiple separate messages, resulting in:

• Excessive notification spam for users
• Poor readability with content arriving in scattered pieces
• Unnecessary API calls to the WeChat platform

For example, a simple 3-paragraph message would be split into 3+ separate messages even if the total length was well under WeChat's 4000 character limit.

Solution

Modified _split_text_for_weixin_delivery() (lines 758-778 in gateway/platforms/weixin.py) to:

1. Prefer single-message delivery - Only split when content actually exceeds the max_length (4000 chars)
2. Remove the and "\n" not in content condition - This was causing unnecessary splits for any content with newlines
3. Use paragraph-based splitting - When splitting is necessary, use _split_markdown_blocks() for intelligent paragraph-aware splitting instead of line-by-line splitting

Changes

- if len(content) <= max_length and "\n" not in content:
+ if len(content) <= max_length:
      return [content]

The function now:

• Returns the entire content as a single message if it fits within 4000 chars
• Only splits when necessary using markdown-aware block splitting
• Preserves the safety mechanisms for oversized blocks

Testing

• Manual testing confirmed that messages with multiple paragraphs are now delivered as a single message on WeChat
• Long messages exceeding 4000 chars still split correctly at paragraph boundaries

Impact

• Improved user experience on WeChat platform
• Reduced notification spam
• No breaking changes - the function still respects the max_length limit

[teknium1]

Merged via salvage PR #7903. Your contribution — clean minimal 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 @sherunlock03 for the contribution!