feat: add comprehensive suite of diagnostic scripts, Metal kernels, and validation reports to optimize Mamba3 backward pass dispatch and S4096 chunked performance

Author: apstenku123

.gitignore

@@ -62,3 +62,22 @@ docs/tilelang_ports/mamba3_path_b_vs_c.diff
 docs/tilelang_ports/mamba3_path_c_lowered.metal
 reports/raw/*
 vbgui/.DS_Store
+secrets/
+*.gh_tokens
+outputs/logs/*
+outputs/*.log
+outputs/megatron*
+outputs/reindexed/*
+outputs/pr_ingest/*
+outputs/verification_report/*
+outputs/dedup_seen.sqlite
+outputs/dedup_seen.sqlite-shm
+outputs/dedup_seen.sqlite-wal
+outputs/shared_dedup.sqlite
+outputs/_validate_conveyor/run.log
+outputs/agent_trajectories/sample_transitions.parquet
+outputs/conveyor/_done.json
+outputs/packed/1024/build_summary.json
+outputs/packed/1024/build.log
+outputs/verification_report/_summary.json
+outputs/verification_report/_samples/_manifest.json