GLM-5.2 agent gets stuck after successful response: unsupported non-text content error 1210

**Describe the bug**

When using a GLM-5.2 agent through Z.AI’s OpenAI-compatible coding endpoint, Buzz successfully receives the initial model completion but then
gets stuck retrying the follow-up request. The agent never visibly replies.

**Steps to reproduce**

1. Open Buzz Desktop on Windows.
2. Use an agent configured with the OpenAI-compatible provider, Z.AI coding endpoint, and model `glm-5.2`.
3. Send the agent a message in a channel or direct message.
4. Observe that Buzz retries the run instead of displaying the reply.

**Expected behavior**

The GLM agent should display its response normally. If the provider does not support the message format, Buzz should show a clear error
instead of retrying indefinitely.

**Version and platform**

- Buzz version: unknown
- OS: Windows

**Logs / additional context**

The initial GLM request succeeds. The follow-up request fails repeatedly with:

```text
400 Bad Request
code 1210:
messages.content.type is invalid, allowed values: ['text']

This appears to happen because Buzz resends conversation history containing a structured content block, such as a tool call, reasoning block,
or metadata, while the Z.AI coding endpoint only accepts plain text content blocks.

Limiting the agent to one round appears to avoid the failing follow-up.

No API keys or private configuration files are included.

Start in the OpenAI-compatible provider's conversation-history and follow-up request handling, then reproduce with the Z.AI coding endpoint and glm-5.2 configuration described in the issue. Trace which structured content blocks are resent after the initial completion; done means the agent displays its response or reports a clear unsupported-format error instead of retrying indefinitely.

Written by the indexing model from the issue text.