reuse the linear operations in quaternions. Also mark some constructors
explicit to better spot coding errors.

function use degrees rather than radians.

cross product, normalize, etc.

at (0,0).

to modify their precision at runtime.

mode, and do not call "new" when we’re not sure "delete" can be called.

build host.

negative zero or negative nth roots are not handled.

multiplications where appropriate.

the 16-to-32-bit refactoring.

functions.

solving.

sin(pi/2-x). No loss in precision in theory.

with large values. Until now they were evaluating the Taylor series even
for huge values.

these functions.

so that their initialisation order ensures they have the proper values. Add
unit tests for these constants.

and simplify the Remez code accordingly.

Remez exchange program.

multiplications by integers, especially powers of two.

accurate.

improved implementation of exp(), which relies on log().