Multiple trajectories for identical training parameter values during operator interpolation

[shanemcq18]

When using any interpolatory operators, all data corresponding to a fixed training parameter value (even if there are multiple trajectories for different initial conditions or input functions) must be grouped into the same "trajectory", otherwise we are providing multiple values for a single parameter and trying to interpolate (see #71). This should probably be detected and fixed in ParametricROM.fit(), but need to be careful not to mess up fit_regselect_*().