docs: Phase 9 cross-domain forgetting matrix (emc)

docs: consolidate bench-vs-base table (P6+P7)

docs: Phase 7 CUDA bench results (emc)

docs: add upstream Qwen3-4B-Instruct-2507 official evals

docs: add base vs tuned bench comparison

docs: standardize Phase 6 bench table (mascarade column)

docs: replace Phase 6 Gemma bench with real Qwen3-4B held-out eval (emc)

docs: add Apache-2.0 license + ailiance-bench v0.2 results

Training in progress, step 210

Training in progress, step 200

Training in progress, step 100

initial commit