Fix/sendmessage response check and tool result truncation

[aalinyu]

Summary

• Parse sendMessage API response body: check ret / errcode and throw on non-zero values, catching silent failures where the HTTP request succeeds but the WeChat backend returns an error.
• Truncate tool result echoes: the deliver callback now identifies kind === "tool" payloads and truncates non-error tool results exceeding 120 characters, preventing large web_search results from flooding the chat.

Root cause

Two issues were observed in the WeChat channel:

1. Some messages return HTTP 200 from sendMessage but never reach the WeChat user. The sendMessage function was discarding the response body, making it blind to backend-level ret / errcode errors.
2. When the AI invokes tools like web_search, raw search results (4,500–8,000 chars each) are echoed as visible chat messages. A single multi-tool turn can emit 17 messages in 48 seconds, burying the AI's final reply under a wall of tool output.

Changes

File	Change
src/api/types.ts	Add ret / errcode / errmsg fields to SendMessageResp
src/api/api.ts	Parse sendMessage response JSON; throw Error when ret !== 0; gracefully degrade on non-JSON responses
src/messaging/process-message.ts	Accept deliveryCtx?.kind in deliver; truncate non-error tool results to 120 chars