CAPIError 400 invalid message content type <nil>

### Short summary

When using a local model (seems to be any model loaded with Ollama), after one response from the model, any subsequent requests will instantly error with the following error: CAPIError 400 invalid message content type

### Affected version or release

v1.0.25

### Installation context

Installed on Windows 11 desktop, private repo (can provide if needed)

### What happened?

Trying to use a local model served by Ollama, getting the following error after one response from the model:

`CAPIError 400 invalid message content type `

Seems to happen with any model loaded in Ollama that supports tool calling - attempted with `gpt-oss:20b` and `empero-ai/Qwable-9B-Claude-Fable-5-GGUF` (from huggingface) models.

I haven't found any relevant errors in any Github issues, nor any relevant search results. There isn't any workaround except for starting a new session, where I can only make one request before getting the error again.

### Steps to reproduce

1. Add local Ollama provider with any model that supports tool calling
2. In Copilot app, make any request in a new session
3. After first response from the model, make another request
4. See error

### Expected behavior

Subsequent requests should be attempted and respond as normal without errors

### Additional context

_No response_

Reproduzieren Sie das Problem in der Copilot-App mit einem Ollama-Modell mit Unterstützung für tool-calling und verfolgen Sie anschließend, wie die zweite Anfrage nach der ersten Antwort zusammengestellt wird. Konzentrieren Sie sich auf den Nachrichteninhalt, der zu nil wird, und überprüfen Sie, dass wiederholte Anfragen normal abgeschlossen werden, ohne die Sitzung neu zu starten.

Vom Indexierungsmodell aus dem Issue-Text verfasst.