fix: preserve pending work during workspace recovery

[eshulman2]

Summary

• build and validate fork-backed replacement clones before touching the existing workspace
• atomically swap in a validated replacement and restore the original if the swap fails
• explicitly mark recreated workspaces so deterministic path reuse cannot masquerade as workspace survival
• rerun implementation/review after replacement instead of marking an unpushed commit complete

Related open PRs

• fix: ensure fork remote in setup_workspace before implementation push #201 covers early fork creation and state checkpointing
• fix: increment retry_count when prepare_workspace fails #196 covers bounded retries after workspace preparation failures
• this PR covers the remaining replacement safety and identity defects

Testing

• focused workspace synchronization, replacement failure, and pending-push recovery tests: 7 passed
• ruff check on all changed Python files: passed