Benchmark suite overhaul: replace shell scripts with sweep.py (#1341)

* Benchmark suite overhaul: replace shell scripts with sweep.py

- Add sweep.py: Python-based benchmark orchestrator that sweeps forward
  or backward across commits, cloning the repo into a temp directory and
  running benchmarks via uv. Results are stored as per-benchmark JSONL
  files on gh-pages with crash-safe commit range tracking.

- Modernize run.py: clone-based execution (like sweep.py) with --commit
  flag for pinning to a specific SHA and --local flag for local checkout.
  Switch from absl flags to argparse, run via uv instead of bare subprocess.

- Remove legacy shell scripts: backfill.sh, nightly.sh, run.sh are fully
  superseded by sweep.py and run.py.

- Update systemd service and README to reference sweep.py.

* address review: refactor CLI, discovery, assembly, and env vars

Author: erikfrey