add use_streaming config to use chat completion streaming

[CasparChou]

Changes Summary

📝 Overview

Adds use_streaming configuration option to enable chat completion streaming mode. Required for LLM providers (e.g., ChatGPT subscription via litellm) where non-streaming API is unavailable.

🔧 Changes

Modified Files: 3 | Additions: +152 | Deletions: -3

1. pkg/config/config.go

• Added UseStreaming boolean field to LLMConfig (default: false)

2. pkg/llm/classifier.go

• New createChatCompletion() method: routes to streaming or non-streaming based on config
• Streaming mode: accumulates streamed chunks into standard API response format
• Updated 3 API calls to use new method

3. pkg/llm/classifier_test.go

• Added TestClassifier_Streaming() - validates streaming functionality
• Added TestClassifier_Streaming_DefaultOff() - ensures backward compatibility (disabled by default)

[CasparChou]

hi @umputun I accidentally set the wrong base branch when opening the PR after forking. I also wanted to check whether you’re interested in this feature. It allows using a litellm proxy to connect to the ChatGPT subscription plan, which is a cost‑saving approach that won’t incur additional token-based fees.