Add resumable CC:Ladder runs (ladder run --resume)

[john-b-yang]

Resume an interrupted codeclash ladder run in place instead of restarting from rung 1.

Usage

• uv run codeclash ladder run <config> --resume <interrupted-run-log-dir>

How it works

• Reuses the interrupted run's log dir and push branch (dir name = branch name).
• Scans per-rung metadata.json to find the first not-yet-cleared rung + the climber's codebase tag at the last cleared rung; skips cleared rungs and seeds from that tag via branch_init.
• First resumed rung force-resets the branch (--force-with-lease) to discard the interrupted rung's partial pushes; later rungs fast-forward.
• Clears the interrupted rung's partial dir so it re-runs cleanly.

Checks (fail fast)

• Requires push: True (pushed branch + round tags are the codebase store).
• Refuses if the run already finished (top-level metadata.json present) or ended on a clean loss / full clear.
• --resume dir must contain a rung matching the config.

Notes

• Zero new bookkeeping in normal runs — resume is derived entirely from artifacts already written (ladder_advancement.cleared, pushed round_tags, and the top-level summary as done-marker).
• Touches 3 files: cli/ladder.py (flag), tournaments/ladder.py (resume logic), agents/player.py (force_push).
• Scan logic verified across scenarios (interrupted-between-rungs, crashed-mid-rung, clean-loss, all-cleared, empty); not yet exercised end-to-end with Docker.