The current implementation assumes that initial guesses (p0) and bounds are scalar, e.g., 0-D variables.

When fitting, e.g., a 1-D function to 2-D data, we may have different initial guess and bounds for each slice of the data, i.e., for each independent fit. curve_fit should be able to support this.

• Make sure to check coords of initial guesses, if present this should match the data's coord (the coord not participating in the fit).
• Initial guess must have dims, just like data and we should reject, e.g., plain NumPy arrays which would require positional association of values with data slices.
• Slice dims for p0 and bounds.
• Note that different params can have a different set of dims.