fix(taskctl): resurrectionScan doesn't handle adversarial-running stage on Pulse restart

[randomm]

Task Description

When Pulse restarts, resurrectionScan iterates over in-progress tasks and handles:

• stage: developing → advances to reviewing (preserve worktree)
• All other stages → reset to idle (destroy worktree)

But adversarial-running tasks are NOT handled correctly. They fall into the "other stages" branch, which destroys the worktree and resets to idle — losing the developer's work. The correct behavior is to reset stage to reviewing (preserving worktree) so Pulse re-spawns the adversarial agent on the next tick.

Root Cause

Line 122 only checks for stage === 'developing'. The adversarial-running case should also preserve the worktree and reset to reviewing.

Fix

Add adversarial-running handling in resurrectionScan:

• Clear assignee / assignee_pid
• Reset stage to reviewing (so adversarial re-spawns)
• Preserve worktree and branch (developer's work is there)
• Add comment: 'Resurrected: adversarial session ended before restart. Returned to reviewing.'

Quality Gates (Non-Negotiable)

• TDD: Write tests before implementation
• Coverage: 80%+ test coverage for new code
• Linting: All code passes project linting rules
• Local Verification: All tests pass locally before completion

Acceptance Criteria

• resurrectionScan resets adversarial-running tasks to reviewing stage (not idle)
• Worktree and branch are preserved
• Unit test covers the new branch in pulse.test.ts or a new fixture