[evandhoffman]

Closes #37.

What changed

• summarize trace behavior into stable per-run counters for assistant answers, verification/self-correction markers, format-fixation markers, duplicates, and total trace lines
• snapshot the workspace so each run summary records created/modified file counts alongside app.py hash, size, LOC, route count, inline-html usage, and dependency count
• add helper coverage plus an end-to-end summary test that writes representative trace and workspace artifacts through _run_one

Verification

• uv run python -m unittest tests.test_runner
• uv run python -m unittest discover tests