Add block transposed memory layout type

Description

To achieve the targeted 2x+ speedup over baseline SIMD for multi-vector distances, the memory layout must be specifically optimized to feed the microkernels efficiently. This tracking issue reflects the integration of BlockTransposed, a new Repr implementation that stores multi-vectors in a layout optimized for SIMD processing. This enables faster multi-vector distance computations by ensuring data is arranged for efficient vectorized access patterns.

Tasks

• Implement the core block-transposed matrix representation to enable contiguous memory loads into SIMD registers.
• Create access abstractions (row proxies and block views) to allow microkernels to seamlessly consume the transposed data.
• Validate layout correctness and memory safety across various packing configurations and data types.