feat(game): stream validation for game metadata

[Seungpyo1007]

What & why

Makes game metadata a first-class validated dataset while keeping the million-record import and its CI practical.

• validates game title, boolean verification status, numeric metrics, classification lists, ESRB values, image URLs, slug uniqueness, and shard paths;
• streams game records for both python -m app.validate and python -m app.validate --category game, rather than retaining record JSON in memory;
• runs the full data validator only for data/** changes and main pushes; validator-only PRs run focused unit tests without checking out the million-record dataset; and
• documents the sparse data workflow, its recovery guardrail, and focused regression tests.

Source

Not applicable: validator and repository-structure work aligned with the existing TechEngine Game model. No TechEngine code or submodule pointer changes.

Checklist

• Focused game validation unit tests pass locally and in CI
• Default validation is regression-tested not to call _load(game)
• Actual la game shard passes streaming validation (1.705s, 2.03MiB peak)
• Full all-category validator remains required for data/** PRs and main pushes
• Full game scan uses streaming record iteration rather than retaining record JSON in memory
• Game paths, slugs, and source URL rules are validated
• Sparse-index recovery avoids assume-unchanged and skip-worktree