DuckDB chunk query blocks async executor

[kylebarron]

lazycogs/src/lazycogs/_backend.py

Lines 337 to 368 in 577205d

	async def _run_one_date(
	t_idx: int,
	plan: _ChunkReadPlan,
	) -> dict[str, np.ndarray] | None:
	"""Read and mosaic all COGs for a single time step.
	Issues one DuckDB query for items overlapping the chunk at this date, then
	calls async_mosaic_chunk to fetch and reproject all tiles.
	Returns None if no items match the query.
	Args:
	t_idx: Index into ``plan.dates`` for the time step to read.
	plan: Read plan carrying all parameters for this chunk.
	Returns:
	Per-band arrays keyed by band name, or ``None`` if no items matched.
	"""
	date = plan.dates[t_idx]
	with plan.duckdb_lock:
	items = _search_items(
	plan.duckdb_client,
	plan.parquet_path,
	plan.chunk_bbox_4326,
	date,
	plan.sortby,
	plan.filter_expr,
	plan.ids,
	plan.filter_fields,
	label=f"bands={plan.selected_bands!r}",
	)

This looks to be a synchronous, blocking duckdb call inside the main-thread async executor?

[hrodmn]

I did some testing and found that the lock that I added to the duckdb client is definitely not necessary since the duckdb client handles concurrent traffic on its own. I don't think #39 really solves this issue but it should make it less bad.