ENH: hyperplane() improvements and calculate() parameter vectorization

[richardotis]

• Adds a obj_parameters_2d function to PhaseRecord, which allows broadcasting an array of degrees of freedom against an array of parameters.
• Rewrites hyperplane for performance to use the system memory allocator, instead of Python, to create arrays. hyperplane is now a nogil function.
• An algorithmic improvement to hyperplane uses the fact that points above the candidate hyperplane at a given iteration can never be on the solution hyperplane. This means we only need to compute driving forces for points below the candidate hyperplane.
• calculate now supports a vector of parameters (dict of equal-length lists) as an argument. An extra dimension is created in the output Dataset to accommodate this. For the case of single-valued parameters, the extra dimension is omitted in the output for backwards compatibility. There's now a data array called param_values with coordinates called samples and param_symbols, where param_symbols is a coordinate with the parameter symbol names as strings

[bocklund]

This looks great.

Beyond just doing vectorized calculations internally for something like ESPEI, this has meaningful implications to giving a faster API to do the post-processing and UQ calculations that are done by PDUQ, given a Markov chain of parameter samples.