Add support for general CSR(A) x CSR(B) sparse matrix multiplication

[Saptami191]

Fixes #21

Summary

This PR adds support for general CSR(A) × CSR(B) sparse matrix multiplication instead of assuming self-multiplication (A × A).

The current implementation in AxBRowIP.c loaded the same CSR files for both matrices and reused the same dimension metadata, which restricted the kernel to square self-multiplication workloads and caused incorrect handling for general rectangular SpGEMM cases.

---

Changes

General SpGEMM Support

• Added separate CSR export/import for matrices A and B
• Added support for independent:
  • dimensions (M, K, N)
  • NNZ counts (aNNZ, bNNZ)
  • CSR arrays for A and B

Dimension & Correctness Fixes

• Fixed incorrect initialization of:
  • N
  • bNNZ
• Updated accumulator/output traversal to correctly iterate across N instead of K
• Added validation support for rectangular matrix multiplication

COO → CSR Improvements

• Added sorting of COO entries before CSR conversion to ensure stable CSR row ordering
• This helps preserve correctness assumptions for future traversal/merge-based sparse optimizations

Test Matrices

Added:

• square matrix examples
• rectangular matrix examples

for reproducible validation and testing.

---

Validation

Tested successfully on:

• square matrix multiplication
• rectangular matrix multiplication

Verified example:

[
(2 \times 3) \times (3 \times 4)
\rightarrow (2 \times 4)
]

Example runtime output:

Matrix A: 2 x 3, NNZ = 3
Matrix B: 3 x 4, NNZ = 4

Estimated cNNZ = 10
Actual cNNZ = 4

Resulting Matrix C (non-zero entries):
C[0][0] = 14.00
C[0][1] = 5.00
C[0][3] = 16.00
C[1][2] = 18.00

---