fix(multimodal): remove fake image support, model cannot decode base64

feat(multimodal): auto-compress large images for reliable processing

feat(multimodal): working image support via base64 text embedding

feat: add tool_choice / toolConfig support for all API paths

feat: full Google Native API support for Gemini CLI compatibility

feat: multimodal framework + image upload (WIP)

fix: code review issues (P1-P3)

feat: modularize + true streaming + multimodal support