Add plog_core_double with fallback for AVX without AVX2

Revert "Speed up plog_double ~1.7x with fast integer range reduction"

Speed up plog_double ~1.7x with fast integer range reduction

Switch ASAN/UBSAN smoketest pipelines to large runners

Improve psincos_double: faster polynomials + accurate range reduction

Fix bugs and improve robustness of SelfAdjointEigenSolver, improve test coverage

Improve dense linear solver docs with practical guidance

Vectorize BLAS level 1/2 routines with Eigen expressions

Replace blas/f2c with clean C++ implementations

Prevent nightly CI pipelines from being auto-cancelled

CI: Add AVX512-FP16 build tests with GCC 13

Fix GCC maybe-uninitialized warning in InnerProduct

Speed up pexp_double by ~15-17%

Speed up plog_float by 1.6x with improved accuracy

ulp_accuracy: use dynamic work queue for thread load balancing