Migrate mixed optimizer to OSS (pytorch#2573)

Summary:
Pull Request resolved: pytorch#2573

Moves `optimize_acqf_mixed_alternating` to OSS BoTorch. This is an optimizer designed for mixed (low-cardinality) integer and continuous variables. It alternates between discrete and continuous optimization steps to optimize the acquisition function. The discrete step greedily searches over the integer variables, moving to the 1-Manhattan distance neighbor that offers the greatest improvement at each step. The continuous steps calls to `optimize_acqf` to optimize the continuous variables, while keeping the discrete dimensions fixed.

Reviewed By: Balandat

Differential Revision: D64349675