Further Changes to Alri (if needed)

[tbhallett]

This builds on the changes in #676 and can be used to make any further changes following the review and work in July 2022.

Things discussed with @inesll, @timcolbourn and @tbhallett are:

• Introduce parameter for the probability with which oxygen is indicated in the case of danger_signs

• Introduce parameter that governs whether oxygen is provided to those with p.o. when O2 < 90%.

This is expected to show that there is a much greater impact of oxygen / pulse oximetry when....

• danger signs is not such a good proxy for needing oxygen.
• oxygen is provided to more people in need of it (i.e. those with saturation 90-92%).

I have been thinking that perhaps both of these could be accomplished by:

1. Changing the line in treatment_failure: needs_oxygen = SpO2_level == "<90%" [I think this should probably change in any case, actually...?].

2. Creating some more classification as follows: danger_signs_pneumonia_oxygen_needed and danger_signs_pneumonia_oxygen_not_needed

3. Elaborating _ultimate_treatment_indicated_for_patient for indicate oxygen under each classification accordingly.

4. Elaborating get_imci_classification_based_on_symptoms so that those with danger_signs among their symptoms get assigned to danger_signs_pneumonia_oxygen_needed and danger_signs_pneumonia_oxygen_not_needed based on something (e.g. cyanosis symptom, or a flip of a coin, based on a parameter). [If a coin-flip would need to save the imci_classification_based_on_symptoms as a property on the data fame so that it's fixed for the episode.]

5. Elaborating _get_imci_classification_by_SpO2_measure so that those with Oxygen < 90 (and <92% when the parameter says so) get assigned to danger_signs_pneumonia_oxygen_needed and danger_signs_pneumonia_oxygen_not_needed based on their actual measurement of SpO2.