Tiled splineTransform! for RLR/SLR and multi-tile stride fix for 3D z-major layouts

[mmbell]

Summary

Distributed support for the 3D spline-vertical grids, driven by the Scythe moist_compressible_RLR equation set (radially tiled RLR runs are needed because the 3D cylindrical grids are computationally expensive):

• Add the 3-argument (tiled) splineTransform! for RLR_Grid — it existed for RZ/RiRk, RL, SL, RLZ and SLZ but not RLR, so any distributed RLR run failed with a MethodError at the first spectral sync. Same per-z_b block layout as the RLZ method, with the spline vertical's b_kDim blocks.
• Add the matching SLR_Grid method (same layout; the spherical spline-vertical grid had no tiled form either).
• Fix the patch strides in the z-major tiled methods (RLZ/RLR/SLZ/SLR): the patch z-level base index was computed with the tile's wavenumber block count. Sub-blocks are aligned by wavenumber, but an inner tile carries fewer wavenumbers than the patch (its kDim follows its outermost ring), so on inner tiles every z level past the first read the wrong shared rows. Single-tile (tile = patch) behavior is unchanged — the strides coincide, so existing single-worker results are bit-identical.

Tests

• New 3D multi-tile splineTransform! testset in test/tiling.jl: a B→A round trip over a 3-way radial decomposition for RLZ, RLR and SLR, seeded with wavenumber-1 content so the k ≥ 1 blocks are exercised. Verified to fail against the pre-fix stride math (the three value checks catch it) and pass with the fix.
• Full suite: 35246/35246 pass.
• End-to-end (Scythe side, companion commit on development): the same moist_compressible_RLR warm-bubble run with 1 worker vs 2 radial tiles for 300 steps agrees to ~4e-15 relative on every prognostic field.

[codecov]

Codecov Report

❌ Patch coverage is 98.52941% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/tiling.jl	98.52%	1 Missing ⚠️

Files with missing lines	Coverage Δ
src/tiling.jl	84.87% <98.52%> (+0.42%)	⬆️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[mmbell]

Added b0f664c: num_columns(::RRR_Grid) returned the spectral count b_jDim * b_kDim instead of the physical vertical-column count iDim * jDim that every other vertical-column grid returns (iDim for RZ/RiRk, jDim for RLZ/RLR/SLZ/SLR). Model drivers stride the physical array by kDim over num_columns, so the old value under-covered the grid; no existing caller depended on it. Needed by the new Scythe moist_compressible_RRR set.