feat: add unload_every_time toggle and fix model loader edge cases

- Add unload_every_time option to QuantFuncGenerate to offload models
  from GPU to CPU after generation (keeps pipeline alive for fast reload)
- Fix QuantFuncModelLoader/AutoLoader mutating cached PipelineConfig output
- Broaden vision_encoder integrity check to all image-edit series, also
  clean up quantfunc_config.json on re-download
- Untrack bin/{linux,windows}/resource_cache.json (already gitignored)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>