sweep: multi-pattern FILES, resume on interrupt, RESET=1

[pruiz]

Summary

Add three usability improvements to the deep sweep runner:

1. FILES Makefile variable — accept comma-separated patterns (e.g. make sweep FILES="src/a.py,src/**/*.cs") alongside the existing single FILE variable.

2. Resume on interrupt — track successfully scanned files in a plain-text state file (tmp/sweep-state.txt). If the sweep is interrupted or a file fails, restarting will skip already-completed files and pick up where it left off.

3. RESET=1 / RESTART=1 — clear the state file and force a fresh sweep of all files.

Motivation

• Sweeping multiple files currently requires multiple make sweep FILE=... invocations.
• If a long sweep is interrupted (CTRL-C, network issue, model error), there is no way to resume — you must restart from scratch or manually track which files finished.
• A plain-text state file (tmp/sweep-state.txt, one path per line) keeps it simple, inspectable, and editable by hand.

Implementation plan

tools/run-sweep.py

• New STATE_FILE constant -> tmp/sweep-state.txt
• Functions: load_completed(), mark_done(file_path), clear_state()
• New CLI args: --files (comma-separated), --reset (clear state)
• main(): split --files into --file args, load/filter completed set, mark after each successful sweep, handle aggregate rollup for resumed runs (pull prior summaries for skipped files)

Makefile

• sweep target: split FILES on ,, pass each as --file, propagate RESET/RESTART as --reset
• Keep backward compat with FILE

docs/file-risk-sweeps.md

• Update examples and document resume/reset behavior

tests/test_run_sweep.py

• Cover state file lifecycle, --files splitting, skip-on-resume, --reset, aggregate rollup with skipped files