Improve the short-term models

[hazarane]

Description

• every time an intervention is switched off first time (eg selected business open now), switch on another 'mock' intervention ('businesses open but with precautions) which will stay on either until end of time, or until the intervention is on again

• extend the forecasts to present, based on data till end of June. Fill in the feature vector with the last known value. When training the model, use just the deaths&cases till end of Jun, and let the model extrapolate (basically fill in Jun with last known state from May, use Jun cases and deaths for training, extrapolate to Jul, and display the Jul cases and deaths as validation)