Add confidence threshold for recovery to avoid false positives

[luigi-agosti]

Problem

When the original element is removed and no good match exists, recovery returns a low-confidence false positive instead of no match.

Benchmark failure

• recovery-006: 'delete button' was replaced with 'Archive', but recovery returns 'Edit' button (score ~0.3)

Proposed solution

Add a minimum confidence threshold for recovery matches. If best match score is below threshold (e.g., 0.5), return no match rather than a false positive.

Impact

Would improve Recovery rate from 87.5% to 100%