Prototype prepared indexing execution for codec consumers

[d-v-b]

🤖 AI text below 🤖

This draft adds an opt-in prepared execution layer on top of zarr-indexing/grid-partition. It lets codec consumers obtain chunk-local and result selectors directly from affine runs or connected-component tables, while retaining declarative projection lowering for other transforms. It does not replace Zarr's default indexers.

Preparation makes snapshot/borrow ownership and read/write conflict policies explicit. NumPy and shard consumers lower the same work to their respective selector layouts. The selection-flow guide remains on the base branch; this draft separately documents the experimental execution contracts and current sharding materialization boundary.

Review corrections included here:

• Codec integration tests live in the package suite and skip only when Zarr is unavailable. The package CI runs them in the repo-root environment. Both pipelines cover v2, v3, and sharded arrays.
• Complete boundary writes use the valid data extent. Storage-read assertions verify that full boundary writes, including stride-2 singleton selections, skip read-modify-write. Reverse selectors remain conservative because the codec shortcut also assumes buffer-order values.
• Orthogonal scalar axes are removed before applying the remaining orthogonal selection. Negative scalars retain literal-coordinate bounds behavior. Vectorized mixed-scalar normalization remains outside this prototype's Zarr compatibility claims.
• The mypy variable-type collision is fixed. Benchmark documentation contains reproduction and interpretation guidance, without historical timing tables or release notes for a private API.

Validation: 1,550 package tests passed, including 40 real codec integration cases; Ruff and commit hooks passed; _execution.py passed mypy; Pyright reported zero errors (95 warnings); strict MkDocs build passed. No end-to-end storage speedup is claimed. Compact sharding consumption and broader dispatch remain future work.