Optimized Matrix Multiplication and Validation Logic

[BehrozKarim]

• Implemented naive, blocked and parallel matrix multiplication
• Updated the readme with performance results
• Added a writeup.pdf explaining the implementation and the optimization journey

naive_matmul: This function is the naive textbook i -> j -> k triple loop with a local sum accumulator.
blocked_matmul: six-loop tile structure with block_size = 32. Inside each tile, inner loops use i -> k -> j instead of the pseudocode's i -> j -> k. This makes the innermost loop access B and C with stride 1, allowing the hardware prefetcher and the compiler's auto-vectorizer to work. std::min(...) handles non-multiple-of-block-size dimensions; C is zero-initialized because each tile accumulates into it.
parallel_matmul: #pragma omp parallel for over the outer i loop. Each thread zeros and then fills a disjoint set of rows of C (so no atomics or reductions are needed), using the same i -> k -> j inner ordering.

We came up with the i->k->j inner ordering trick because we were not achieving any speedups with the pseudocode version of the implementation.