From kwargs benchmarks and improvements

[moeyensj]

• 265e1b3 (Add from_kwargs benchmarks)
• 4297dad (Add a fast path to Column._load when data is a pyarrow chunked array or standard array and the dtypes match)
  • Enhanced Column._load to skip Arrow cast when data.type == self.dtype.
  • Impact: Cuts unnecessary casts for flat columns during from_kwargs, reducing Arrow compute overhead observed in profiles.
• 5f412fa (Implement SubTableColumn._load with fast paths to avoid casting when dtypes match)
  • Added SubTableColumn._load to normalize inputs into StructArray:
  • Accepts matching quivr.Table via to_structarray(); ensures nullable fields when needed.
  • Returns Arrow StructArray/ChunkedArray as-is if dtype matches; otherwise casts once.
  • Impact: Reduces casting/allocations for nested tables created via from_kwargs, improving constructor performance in nested workloads.

AI-generated summary of AI-generated code.

Benchmarks at 265e1b3:

Benchmarks at 4297dad:

Benchmarks at 5f412fa: