Handle PhaseShiftingTransformers in Parallel

[m-bossart]

We currently skip parallel reduction of phase shifting transformers because depending on the parameters, we cannot necessarily derive an exact equivalent representation in the Ybus:
https://github.com/NREL-Sienna/PowerNetworkMatrices.jl/blob/bc5a33ce6de08b3ee3e84934630287e855348ee2/src/definitions.jl#L13

However, with the current design, if we have parallel phase shifting transformers, we cannot have two entries in the direct_branch_map with the same key, therefore we over-write entries in add_to_branch_maps!:
https://github.com/NREL-Sienna/PowerNetworkMatrices.jl/blob/bc5a33ce6de08b3ee3e84934630287e855348ee2/src/Ybus.jl#L199

[m-bossart]

@jd-lara @SebastianManriqueM
The reason we skip the parallel reduction of phase shifting transformers is we are not able to back out equivalent branch parameters for the PowerModels interface if we have parallel transformers with different angles (the angle has a real component): https://github.com/NREL-Sienna/PowerNetworkMatrices.jl/blob/f4d9e2a815bd15b8abec9a420494474c0887de72/src/common.jl#L315-L320

One option is for get_equivalent_physical_branch_parameters to return a Set{EquivalentBranch}. Then for these tricky cases, we can return two EquivalentBranch components representing the parallel transformers. We can then allow these parallel combinations to be reduced, and bypass the over-writing of entries described above.