visual-compare: support comparing against previous run baselines

## Summary
`wordpress.visual-compare` should support comparing a current result against a previous visual comparison run or declared baseline artifact.

## Problem
Callers often need to know whether a current change improved, regressed, or stayed neutral relative to earlier evidence. Today that comparison requires external bookkeeping and custom result aggregation.

## Proposed shape
- Accept a previous `visual-diff.json` or aggregate visual comparison artifact as a baseline.
- Emit deltas for mismatch ratio, mismatch pixels, dimensions, and status.
- Support both single-comparison and matrix comparison outputs.
- Keep policy decisions outside the command; report evidence only.

## Acceptance criteria
- A visual comparison run can report deltas against a previous run.
- The output clearly distinguishes current source/candidate comparison from baseline delta comparison.
- The command does not enforce pass/fail policy beyond command execution success.