Add benchmarking suite for multi-vector operations

[suri-kumkaran]

Add benchmarking suite for multi-vector operations

Description

Establish a rigorous benchmarking framework to validate the performance and scalability of the new multi-vector architecture. This suite will measure the standalone API's throughput, latency, allowing us to compare our specialized microkernels and memory layouts against established baselines as they are introduced.

Tasks

• Set up the core benchmarking harness for the standalone multi-vector distance API.
• Integrate naive or external baselines for comparative performance tracking.
• Benchmark the core distance functions across supported datatypes (e.g., f32, f16) and quantizations.
• Establish and document the methodology for tracking future hardware-specific microkernel optimizations.