feat: harness SWE-bench into AISO Coding Agent stack (HF Datasets + HF Inference API)

feat: add SWE-bench adapter verification results (2,218 tests across 296 files)

feat: real functional benchmarks — AISO WASM parity, ScrollCast tamper detection, motif throughput, latency histograms