Fix config.json: remove ghost embedding_file + add per_layer_embedding_file (Cycle 10 ship-bug)

Remove legacy PLE-schema files (taobao schema is canonical)

Cycle10 taobao-schema re-export 2026-04-25

Sync verified TokForge 3.4.9 Gemma 4 MNN bundle

V2 (FusedAttention disabled + force_full_decode_recompute): llm.mnn.json

V2 (FusedAttention disabled + force_full_decode_recompute): llm_config.json

V2 (FusedAttention disabled + force_full_decode_recompute): llm.mnn

V2 (FusedAttention disabled + force_full_decode_recompute): export_args.json

V2 (FusedAttention disabled + force_full_decode_recompute): llm.mnn.json

V2 (FusedAttention disabled + force_full_decode_recompute): llm_config.json

V2 (FusedAttention disabled + force_full_decode_recompute): llm.mnn

Add llm.mnn.weight (Q4+int4 PLE, requires TokForge 3.4.9)

Q4 weights + int4 PLE (requires TokForge 3.4.9)

Add per_layer_embeddings_int4.bin (Q4+int4 PLE, requires TokForge 3.4.9)

Add embeddings_int4.bin (Q4+int4 PLE, requires TokForge 3.4.9)

Add llm.mnn (Q4+int4 PLE, requires TokForge 3.4.9)

Add tokenizer.txt (Q4+int4 PLE, requires TokForge 3.4.9)

Add config.json (Q4+int4 PLE, requires TokForge 3.4.9)

Add llm_config.json (Q4+int4 PLE, requires TokForge 3.4.9)

Add README.md (Q4+int4 PLE, requires TokForge 3.4.9)

initial commit