chore(publish): prepare repository for public release

[acorn421]

Summary

• Restructure top-level docs and README to serve as the canonical public entry point; mark dashboard as experimental
• Reorganize benchmark suites under category subdirectories (afc/, asc/, atlanta/, crsbench/, smoke/) and split smoke suites by language
• Reset experiment-configs to a clean reference set (local/, gcp/, agentic-cli/, smoke-testing/, discovery/); drop legacy and .bak snapshots
• Remove internal-only artifacts: design docs, deprecated migration module, legacy CRSes, snapshot examples, smoke/debug skills, re-eval/CPV-assignment scripts
• Bundle a quick-start experiment-configs/smoke-testing/first-run.yaml and link it from the README; runs the sanity suite (3 trial jobs) with 3 parallel workers and no LLM credentials
• Fix stale --benchmark-suite sanity / crsbench-all references in docs and CLI help; bump sample max_total_time to satisfy the phase-sum validator; add git submodule update --init --recursive and HF_TOKEN .env notes to Quick Start
• Default RTS/incremental-build off for bug-finding to match the public Tier-1 protocol
• Refresh oss-crs submodule and uv.lock

Test plan

• CI green (lint/typecheck/tests)
• Clean-clone path works: git clone -> git submodule update --init --recursive -> uv sync -> cp .env.example .env -> uv run crsbench prepare + prepare --coverage
• uv run crsbench download --benchmark-suite smoke/sanity succeeds after HF dataset approval
• uv run crsbench worker --experiment-config experiment-configs/smoke-testing/first-run.yaml spawns 3 workers; the matching run completes the 3 trial jobs