Improve the detection of past peak incidence

[thibautjombart]

The current definition of the past peak incidence is 'the highest weekly incidence in the data excluding the last 90 days'. The exclusion of the last 90 days is justified by the fact that we want to avoid defining the peak using data from the ongoing wave, in which case things will always level at 100% (with the peak being last day's incidence).

Problem: some countries may have faster waves and actually have had their last peak in the last 90 days. We need a better approach for detecting past peaks (likely relying on a genuine peak i.e. increase -> decrease rather than a maximum incidence. I think @mv2405 had an approach for this, which we may want to try?

[thibautjombart]

Solution is no longer exclude recent data points from the peak definition, which because trivial