Bug fix: Ensure correct data retrieval for global cell order in dense reader, avoiding fill values

[kounelisagis]

slab_dim is being calculated incorrectly in DenseReader::cell_slab_overlaps_range, causing global cell order reads to return some fill values. After this fix, slab_dim remains unchanged for row-major or column-major requests, but for global order requests, cell order is used to determine slab_dim.

Let's align the calculation with the approach already used in other cases:

TileDB/tiledb/sm/query/readers/dense_reader.cc

Lines 2319 to 2320 in 87c9860

	if (layout_ == Layout::ROW_MAJOR ||
	(layout_ == Layout::GLOBAL_ORDER && cell_order == Layout::ROW_MAJOR))

The initial issue was also verified with TileDB-Py, and after applying this fix, the results are as expected.

[sc-60301]

---

TYPE: NO_HISTORY | BUG
DESC: Fix incorrect slab_dim calculation in DenseReader::cell_slab_overlaps_range to prevent fill values in global cell order reads.

[rroelke]

Checking my understanding:

layout_ is the user's requested data order.

A "slab" is a run of cells within a tile which are contiguous in the users' requested data order. e.g. if the user requests global order then the whole tile is a single slab; if the user requests row-major order then each row within a tile is a slab if the cell order is row-major, etc.

slab_dim is meant to be the dimension which the cells of the slab are contiguous. If the requested order is row major then the slabs are contiguous columns (i.e. dimension 1), if the cell order is column major then the slabs are contiguous rows (i.e. dimension 0).

In your example, the tile and cell orders are column major. The read query is global order, so the whole tile is a slab, so the slab_dim should be chosen using the schema's cell order.

And that is what your fix does: slab_dim is unchanged if the user requests row or column major, but if they request global order then the cell order is used to determine slab_dim.

Without this, slab_dim is incorrect which results in returning an incorrect intersection, leading to fill values used for other cells.

Do I have this right?

[kounelisagis]

@rroelke , my understanding is exactly the same.

[rroelke]

Thanks!