Refactor dense ECS storage and align render hot paths with Bevy

[Milky2018]

Summary

This branch combines the dense ECS table-column refactor with the performance and correctness work needed to use it in Bevy-shaped render hot paths.

• replace optional-slot component tables with dense typed columns, one-shot ownership transfer, direct required-value fetches, and table-batched query traversal
• migrate visibility, transform, sprite, camera, picking, asset, time, and PBR hot paths to the validated batch/value APIs
• harden batch safety and ECS semantics for filtered rows, escaped mutable views, table-0 entities, same-reference replacement, despawn membership, and change detection
• restore source-level render correctness for Camera2d picking/frusta, affine sprite transforms, dynamic shadow phase resolution, frame-local render storage, and PBR material capacity
• upgrade the window/wgpu/WGSL dependency stack, use typed surface handles, synchronize pinned Bevy shader/assets, and refresh accepted visual references
• update the Markdown issue tracker with performance experiments, API migrations, resolved toolchain blockers, and remaining external limitations

Why

The previous ECS representation stored component columns as optional per-row slots and paid repeated wrapper, lookup, and change-tracking costs in render-critical queries. Several early optimizations exposed deeper semantic hazards around ownership, filtered write authority, table migration, and change detection. This branch makes the storage/query boundary dense and explicit, then migrates only validated performance-sensitive consumers.

The render follow-ups fix correctness gaps found during representative Bevy example captures rather than introducing visual heuristics or shader workarounds.

Impact

Performance-sensitive systems can traverse dense table columns in batches while retaining explicit read/write and change-tick semantics. Existing public API migrations and behavior changes are documented in the linked ISS records and generated interfaces.

This is intentionally a draft because the branch is large and should receive focused review across ECS storage/safety, visibility and transform propagation, render/PBR lifecycle, and dependency/binding changes.

Validation

• git diff --check origin/main...HEAD
• moon test utils/utils_test.mbt --target native --release --deny-warn — 4/4 passed on the current default native backend
• standard native frame-90 captures for examples/3d/transmission and the clearcoat multi-layer material path complete reproducibly after the WGSL 0.17 upgrade
• representative 2D/3D example and performance captures are recorded in docs/issues, including many animated sprites, visibility, picking, shadows, PBR, and asset parity work
• examples/3d/meshlet now compiles and captures without the former Moon compiler ICE; exact meshlet output remains blocked and is explicitly tracked by ISS-020

Known remaining blockers

• ISS-020: exact Bevy meshlet visibility-buffer runtime still requires native R64Uint atomic storage-texture support
• ISS-427: Homebrew ASan still deadlocks before main on this macOS host, while Apple ASan does not support LeakSanitizer
• ISS-443: unrelated current-toolchain type-parameter method warnings still prevent treating a repository-wide --deny-warn run as green