update sparse computation, add initial version of BlockCSR and BlockELL

[chaoming0625]

This pull request introduces significant changes to the brainunit/sparse module, including the addition of new sparse matrix formats (BlockCSR and BlockELL), their corresponding benchmarks, and updates to existing imports. The most important changes are outlined below:

New Sparse Matrix Formats:

• brainunit/sparse/_block_csr.py: Added the BlockCSR class, which represents a unit-aware Block-CSR sparse matrix, along with methods for validation, conversion to dense format, matrix multiplication, and sampling sparse matrices.
• brainunit/sparse/_block_ell.py: Added the BlockELL class, which represents a unit-aware Block-ELL sparse matrix, along with methods for validation, conversion to dense format, matrix multiplication, and sampling sparse matrices.

Benchmarking:

• brainunit/sparse/_block_csr_benchmark.py: Added a benchmarking script for BlockCSR matrices to measure the performance of sparse-dense matrix multiplication using different methods.
• brainunit/sparse/_block_ell_benchmark.py: Added a benchmarking script for BlockELL matrices to measure the performance of sparse-dense matrix multiplication using different methods.

Import Updates:

• brainunit/sparse/__init__.py: Updated imports to reflect the new module structure, including the addition of SparseMatrix and changes to import paths for COO and CSR classes.