Ollama silently drops tool calls to functions missing from the tools list

When a model served by Ollama calls a function absent from the request's tools list, the call is discarded and the reply comes back empty, with no content and no tool_calls. The response still reports generated tokens through eval_count, showing the model produced output that was thrown away. In tests on Ollama 0.40.8 with qwen3:0.6b, eight of eight replies were empty when the needed tool was left out, while adding it to the list produced the expected call every time. The behavior stems from Ollama's generic, template-based tool parser, which affects models without a dedicated parser, including qwen3. The suggested workaround is for clients to list every tool the model might call, and the issue is tracked upstream as ollama/ollama#19187.