fix: apply before_each reset-only hooks for static workspace evals

[christso]

Summary

workspace.hooks.before_each.reset is parsed and documented, but reset-only hooks are ignored during normal eval execution. In static workspaces this leaves the shared git baseline and prior run state in place, so file_changes can diff against stale committed content from a previous run.

Repro

1. Configure an eval with workspace.path and hooks.before_each.reset: fast.
2. Run the eval twice against the same static workspace.
3. Inspect the workspace git history and the grader file_changes payload.

Expected

A reset-only before_each hook should actually reset the workspace before the agent run, and file-change capture should compare against a fresh baseline for that reset state.

Actual

before_each.reset has no effect unless a script/command is also present. Static workspaces accumulate agentv-baseline commits and later runs can produce diffs relative to stale prior-run state.

Notes

Observed against a real eval that writes a CSV artifact into a static workspace path on Windows.