Scale finalizer dirty probe for large managed worktrees

[chubes4]

Problem

workspace worktree finalize still hardcodes a dirty-probe budget that is too small for a clean, large managed worktree. On Data Machine Code 0.56.7, finalization returns:

Error: Worktree finalization dirty probe failed: Process command timed out after 5 second(s).

The checkout is clean, pushed, and already has an open PR. Worktree creation now scales its target-tree inspection correctly, but the final metadata handoff retains a separate five-second Git status deadline.

This is a bounded regression in the lifecycle class previously tracked by #964. The command fails safely and does not alter the checkout, but PR metadata is not persisted.

Acceptance criteria

• Finalizer dirty probes use a filterable/default budget appropriate for large repositories.
• A clean large-worktree fixture completes finalization within a deterministic bound.
• Timeout diagnostics retain the failing phase and configured budget.
• Existing targeted get/finalize boundedness from Targeted worktree get and finalize commands hang on a clean managed worktree #964 remains intact.

AI assistance

OpenAI GPT-5.6 Sol via OpenCode reproduced the finalizer timeout after creating and pushing a clean managed worktree, checked prior trackers, and drafted this regression report under Chris Huber's direction. Chris remains responsible for the issue and follow-up.