Support bounded multi-row GPU gray dispatch

[rsasaki0109]

GPU RGB-to-gray previously dispatched every flat workgroup on the X axis: 4096x4096 requires 65,536 groups, exceeding the requested 65,535-group limit despite valid texture dimensions. Keep the original linear shader for single-row dispatch and distribute larger workloads across a balanced rectangle of 256-lane workgroups. Reject padding groups before index multiplication to prevent u32 wrap.

Validate runtime, RGB channels, dimensions/pixel count, existing compute/binding limits and profiler capacity before allocation or cache initialization. Linear and grid pipelines initialize independently in one runtime-owned family, sharing the original layout/16-byte uniform. BT.601 bytes, gray metadata, one pass, transfer/query receipts and public signatures remain unchanged; no scratch or payload copy is added.

Validation:

• Six new tests check every one of the 16,777,216 possible RGB values through the public large-image path (67,108,864 exact RGBA bytes), plus 120 forced-grid mappings (3,655,572 bytes) against independent CPU and frozen linear outputs. A device probe verifies 96 index/validity words at u32 boundaries and padding groups without giant textures.
• Invalid limits/ownership/channels/capacity, independently lazy kernels, strided ROI, concurrent cold calls, live outputs and texture-pool reuse/budget are covered.
• 148 native image/point GPU regressions pass; strict default/all-feature Clippy, two WASM compilation profiles, formatting and diff checks pass. Browser execution is not claimed.
• Seven release resident conditions with two warmups and twenty rounds; every output is checked. The frozen shader shares the current host executor, so this is not historical API-overhead measurement. Unsupported old large-image dispatch has no baseline timing. All supported controls and the new capacity measurement are published; no speedup is claimed.

Implementation and full diagnostic note.

Based on merged PR #145 (main 95fb8e8). Merge is gated on the frozen source head, all 114 named checks succeeding, no pending/failed checks and a clean merge state.