feat: add lumo-spec markdown checks and tap-target hit areas

[OneXeor]

Summary

Combines and reworks the previous spec-check and tap-target work into one OSS-ready PR.

• Adds lumo-spec Phase 1: Markdown source, Anthropic-compatible LLM round-trip, evidence validation, replay support, JSON/text output, examples, and docs.
• Updates lumo-theory tap-target logic to distinguish visible glyph bounds from optional hit_w / hit_h touch containers.
• Documents optical alignment and icon+label tautology review rules in the Lumo skill.
• Rewrites project-specific spec-check language into generic OSS examples (APP-1234, representative Jira/Confluence checks, no internal helper paths).

OSS audit

• New commit authored with public repo identity: Viktor Savchik <onexeor@gmail.com>.
• New PR body and commit message do not include internal project names, domains, or local helper paths.
• Diff scan checked for project-specific names and obvious secret patterns. Remaining token/API-key mentions are env var documentation only.

Tests

• .venv/bin/python -m ruff check tools/lumo/spec tools/tests/test_spec.py tools/lumo/theory tools/tests/test_theory.py
• .venv/bin/python -m pytest tools/tests/test_spec.py tools/tests/test_theory.py
• .venv/bin/python -m pytest tools/tests

Supersedes #1 and #3.