Peak fit: expose parameter uncertainty

[stuart-cls]

It would be good to expose the uncertainties in best fit parameters.

The lmfit ModelResult already contains this information (for actual fit parameters), so it's just a matter of how best to expose it. One proposal is to modify the Fit Parameters output:

g0 area, g0 area err, g0 amp, g0 amp err

This might also be a good time to reorganize that output into reduced features (fit parameters output) and related meta-output. The motivation here is that the Fit Parameters output should not contain redundant / extraneous information for down-stream modelling. Something like

Features	Metas
g0 amp / g0 sigma / g1 amp / g1 sigma	g0 area / g0 area err / g0 amp err / g0 sigma err / ... / chi squared

[stuart-cls]

The calculated area == amplitude for most peak models[1] . So it is either redundant or if the data cuts off before the fitted peak, wrong:

(Green peak here underestimates the peak area by ~20% because the area is not evaluated outside the data window)

I'd like to remove it, but will need some backwards-compatibility so that we don't break old workflows. Options:

1. Break old workflows by removing area feature, signalling to the user that something has changed
2. Keep the feature, but stop calculating the area with np.trapz and silently replace with amplitude value
3. Keep the feature, somehow calculating the np.trapz value only for old workflows

I'd like to include this in #801 or at least in the same release so that we only change the output of the widget once.

[1] Amplitude is not normalized for Moffat, Breit-Wigner-Fano, Damped Harmonic Oscillator Amp (but not the DAVE one)