Update /generate.py to handle "thinking" models

[srdas]

Fixes Issue #1315

In models that return an entire chain of thought, /generate fails as the notebook file name becomes excessively long, as noted in the issue cited above.

The title returned contains all the thinking about the title, including proposed title text buried within qoutes inside the chain of thought (CoT). This modification extracts the suggested title in the CoT and uses it for the title.

The modification also adds a generic failsafe title in case the CoT fails to return a title. For non-CoT LLMs, the code does not pursue this special handling.

To test:
[1] Set chat model provider to Ollama and the model to huggingface.co/lmstudio-community/DeepCoder-14B-Preview-GGUF:Q6_K
[2] Try the following chat command: /generate write "hi" in the file "allo.txt" in python (this will take some time to run, as the model works through its process of reflection).

As shown:

Result:

Additional tests undertaken:

• The chat command /generate add two numbers in python with the same LLM gives the notebook title: Adding Two Numbers in Python - Basic to Advanced.ipynb.
• The chat command /generate write "hi" in the file "allo.txt" in python generates a title: Write 'hi' to allo.txt.ipynb. (Here the LLM is deepseek-chat with provider OpenRouter.)