Split all semicolon-separated statements onto individual lines
throughout the Triton SVD kernels (_svd2_kernel, _svd3_kernel).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Log info message when Triton is not available so users know
fused N=2/N=3 kernels are disabled.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Fused Triton kernels for N=2,3 (3850-5488x faster than torch),
Gram-eigh hybrid for N>=4, Newton-Schulz inverse sqrt, and
subspace-preserving Procrustes alignment.
From PR #4.