Add support for Ollama's think: false parameter for AI suggestions

Description When using a thinking-capable model as the LLM backend via Ollama, AI suggestion requests time out. The model spends thousands of tokens on internal reasoning before producing a title and a few tags. Observed on paperless-ngx 3.1.3 with Ollama and gemma4:12b, GPU-resident, generating at a sustained 40 tokens/sec: Suggestion requests reach 3400+ generated tokens and are still running when the request is cut at exactly 2m0s Ollama returns 500, the reverse proxy returns 504 The same model on the /api/generate endpoint completes in 2 to 7 seconds Raising LLM Request Timeout and the proxy read timeout makes suggestions complete, but each one takes minutes, which is not workable for interactive use. Solution you'd like Pass think: false in the request body when the LLM backend is ollama, for suggestion requests. Ideally exposed as a setting in AI Settings so users who want reasoning for document chat can keep it while disabling it for suggestions. Alter…