赵月辉 / fastgpt-migrated

* feat: support multimodal LLM inputs

- add model capability config for vision, audio, and video inputs
- wire multimodal flags through simple app, chat agent, and workflow LLM nodes
- normalize uploaded files and file links into image/audio/video request parts
- adapt chat, agent, and toolcall runtime context for multimodal messages
- add capability tags, i18n entries, design docs, and related tests

* perf: code

* test: update agent file prompt assertion

---------

Co-authored-by: archer <545436317@qq.com>