Fix evaluator threading logic

Fixed Evaluator's threading logic using std::atomic.

Default threading behavior changed.

By default, let the algorithms use all the available # of threads.

Improved SLIM

• Improved SLIM convergence check logic
• API Change for optimizers
• Include user-user similarity-based recommenders

C++ implementation of SLIM

• Workflow fix
• Implementation of SLIM using std::thread

Windows binary support

Build wheels for Windows platforms.

Fix build environment.

Fix the environment variable (CFLAGS) for cibuildwheel.