Modularize GraphRAG smoke runner

[yvette-carlisle]

Summary

• Keep scripts/graphrag-docker-smoke.py as the command-compatible entrypoint.
• Split the 1470-line GraphRAG smoke runner into focused Python modules under scripts/graphrag_smoke/: context, models, common helpers, benchmark scoring, corpus generation, runtime commands, table mapping, generated artifacts, and runner orchestration.
• Preserve generated manifest/fixture/report behavior while reducing the largest module to 474 lines and the entrypoint to 10 lines.

Batch size

• 11 files changed, 1539 insertions, 1461 deletions, 3000 total changed lines.
• Original runner reduced from 1470 lines to 10 lines.

Validation

• python3 -m py_compile scripts/graphrag-docker-smoke.py scripts/graphrag_smoke/*.py
• no-live smoke: ELF_GRAPHRAG_SMOKE_RUN=0 ELF_GRAPHRAG_SMOKE_ALLOW_HOST=1 ... scripts/graphrag-docker-smoke.py
• RUSTC=/run/current-system/sw/bin/rustc cargo nextest run -p elf-eval --test real_world_job_benchmark graph_rag
• RUSTC=/run/current-system/sw/bin/rustc cargo make fmt-rust
• RUSTC=/run/current-system/sw/bin/rustc cargo make check (403 passed, 92 skipped)