v1.0.2: Extracted step size initialization to a class and improved trust regio

First full release of the torch_numopt package.

v0.3.1

• Added the AdaHessian algorithm.
• Updated documentation with API section for easy access to relevant classes.
• Refactored base classes and moved gradient computation to LineSeachOptimization class.

v0.3.0

-Added interpolation and bisection based line search.
-Added several learning rate initialization methods.
-Removed the AGD algorithm in favor of a damping option in Newton's method.

v0.2.2

Added SGD with line search and Conjugate Gradient.

Skipped v0.2.1 since there were some issues with the metadata of the package which had to be fixed.

v0.2.0

The library is now called torch_numopt so that new algorithms can be implemented.

The existing algorithms have also been optimized.