[FEATURE SUPPORT] Add Triton GEMV kernel and performance tests

[LoserCheems]

Summary

• Introduces a Triton GEMV kernel for efficient matrix-vector multiplication and adds performance tests comparing various implementations.

Root Cause

• The need for an optimized GEMV operation on GPU prompted the development of this kernel.

Changes

• Implemented an autotuned block-tiled GEMV kernel using Triton.
• Reformatted code for improved readability.
• Added parameterized performance tests for GEMV across different frameworks.

Reproduction

• Not applicable as this is a new feature addition.

Tests

• Added performance tests for the GEMV kernel covering CUDA, MPS, and multiple data types.

Compatibility

• No migration concerns or backwards compatibility issues noted.

Checklist

• Linked issue provided ([FEATURE REQUEST] gemv Triton kernel implementation #22 [FEATURE REQUEST] gemv test coverage and fixtures #24)
• Adds or updates tests
• Updates docs if needed
• No perf regressions