v0.5.16

v0.5.16

sorry this release took longer than it should have

• Escape Hugging Face model IDs, GGUF filenames, and quantization types in generated run and snippet scripts
• Keep synthetic GGUF recommendations tied to direct quantizations of the selected checkpoint
• Reject fine-tunes, merges, conflicting lineage, and artifacts without explicit provenance
• Split benchmark fetching, caching, indexing, lineage, and lookup into focused modules
• Lock the development tools used locally and in required CI checks

Thanks @RaghavRD for the benchmark refactor and @hannibal-lee for the generated-script escaping fix.