Add deterministic professional-visual benchmark harness

[rroojrooj]

Summary

• Add an isolated, deterministic hardware-Electron visual capture harness with fixed camera suites, strict input provenance, renderer diagnostics, performance sampling, and atomic evidence publication.
• Add three-course/24-frame baseline coverage, PNG comparison metrics and diff reports, plus visual:capture, visual:smoke, visual:compare, and visual:perf workflows.
• Add the professional-visuals program specification, implementation/test plans, operator guide, and an evidence-backed assessment of the current rendering quality.

Why

Visual changes could not be judged reliably because camera state, asynchronous loading, post-processing, renderer capability, and procedural noise were not controlled as one reproducible system. This PR establishes a stable before-state and objective regression workflow before the later SP-01+ rendering improvements.

Impact

• Normal gameplay remains on the existing live render path; capture APIs are query-gated.
• Benchmark evidence now fails if suite bytes, selected course inputs, Git state, course identity/revision, font settlement, or shader compilation drift.
• The harness is ready, but the recorded visual verdict remains professional visuals NOT READY. The largest debts are terrain/world edges, aerial/material seams, flat surface masks, vegetation/shadows, and course-specific identity.
• The qualifying performance route here is 1280x720; this PR does not claim the later 1920x1080 release gate.

Validation

• npm test - 375/375 passed.
• Focused capture/readiness review suite - passed.
• Clean hardware Electron smoke on RTX 3060 - passed with deterministic frame hash, fonts ready, shader compilation ready, zero fatal events, and zero page-console errors.
• Two three-course baseline runs - 24/24 frames byte-identical.
• 60-second shown-window performance route - completed with valid CPU/GPU timing evidence.
• Independent final code review - Ready, no remaining findings.