Support configurable max_tokens

[OpenTester007]

Currently max_tokens is hardcoded to 4096 in callAPI (popup.js:406). Different models have different optimal values — reasoning models need headroom for reasoning_content, while chat models can work with lower values for faster responses.

[OpenTester007]

Add a max_tokens input field to the Settings page
Default: 4096
Read from storage in initAPIConfig / saveAPIConfig
Pass into callAPI request body instead of the hardcoded value