Performance refactor of v6.0

[vkomenda]

Functional changes

• Replaced a redundant computation M · 1/M in ReedSolomon::new with the identity matrix when the encoder matrix is constructed.
• Pruned and streamlined ReedSolomon::reconstruct leading to reconstruct benchmark improvement.
• Swapped row-column accesses during coefficient application in ReedSolomon::code_some_slices to improve cache locality and to reduce the number of conditional jumps.
• Simplified the implementation of Gaussian elimination while maintaining functional equivalence.
• Within Gaussian elimination, used an asymptotically faster safe Rust primitive for row swapping.

Interface changes

• Crate dependencies have been updated.
• Some previously pub functions in matrix.rs are made private because they are not used outside of that module.