feat(binar): add sparse ↔ dense BitMatrix conversion

[jmbr]

Summary

Add methods for converting between dense BitMatrix and sparse
column/row representations via IndexSet, and expose them in the
Python bindings.

Motivation

BitMatrix has no way to construct a matrix from sparse column or row
descriptions, or to decompose one into its column/row supports.
Consumers that work with sparse GF(2) matrices (e.g., syndrome
indicators, check matrices) must set bits one at a time.

Changes

Rust API (AlignedBitMatrix + BitMatrix):

• from_sparse_columns(&[IndexSet], row_count, column_count) -> Result<Self, SparseConversionError>
• from_sparse_rows(&[IndexSet], row_count, column_count) -> Result<Self, SparseConversionError>
• sparse_columns(&self) -> Vec<IndexSet>
• sparse_rows(&self) -> Vec<IndexSet>

Both constructors take an explicit (row_count, column_count) shape
(matching the scipy.sparse convention), allowing trailing zero
rows/columns. Invalid inputs return SparseConversionError.

Python API (BitMatrix):

• from_sparse_columns(columns, row_count, column_count)
• from_sparse_rows(rows, row_count, column_count)
• sparse_columns() -> list[list[int]]
• sparse_rows() -> list[list[int]]

Invalid inputs raise ValueError.

Testing

• Proptest round-trips for both sparse_columns and sparse_rows
• Trailing zero rows/columns
• Error cases: too many entries, index out of bounds
• Edge cases: empty matrix, identity matrix
• cargo clippy-all clean, cargo fmt clean, mypy.stubtest clean