[ML] Expose anomaly job level parameter to control how quickly modelling will adapt unpredicted values

There is a tradeoff adapting to changes in data characteristics and fitting extended anomalous periods. We try hard to provide good defaults for this to

1. Avoid polluting the model with anomalous periods,
2. Adapt to changes where appropriate without creating an extended period where the model is significantly in error.

We also provide a mechanism for manually excluding unusual time periods after the fact. But if one has many jobs to manage this process can be labour intensive.

This issue is to consider how we can expose a single job level parameter to allow people to tweak the default behaviour to make the model more or less adaptable after some initial learning period. This would likely affect various underlying parameters such as how aggressively we down weight outlying values, how quickly we detect change points, how quickly we age out old data, etc.

Iniziare esaminando i valori predefiniti di modellazione degli anomaly job e i comportamenti sottostanti indicati nell’issue: ridurre il peso dei valori anomali, rilevare i punti di cambiamento e far invecchiare i dati meno recenti. Determinare come un singolo parametro di adattabilità a livello di job potrebbe controllare questi comportamenti dopo il periodo di apprendimento iniziale e definire il compromesso previsto tra l’adattamento ai cambiamenti e l’adeguamento ai periodi anomali.