These sub-sub-namespace typedefs are only used internally; some of them cause problems by creating a form of circular dependency. For example, ol.style.Atlas is created in style/atlasmanager.js; if ol.style.Atlas.Block is in a separate file, ol.style.Atlas is dependent on it, so the new file must precede ol.style.Atlas. However if it precedes it then it has to create the ol.style.Atlas namespace which should be created by atlasmanager.js. To get round this, these typedefs are renamed to remove the sub-sub-namespace.
Fortunately they are all non-api, so the rename should not affect anything.
Instead of having `ol.events.Key` be a listener object or an array of listener objects, it should be less error prone to have it just be a single listener object.
To avoid using too many functions with multiple return types, the `ol.events.*` functions for registering and unregistering listeners no longer accept an array of event types (and only a single key is returned when registering).
To make it convenient for users to register multiple listeners at once, the `observable.on()` method accepts an array of event types. Internally in the library, we should use the less risky `ol.events.listen()`.
Geometry index won't be updated correctly if feature geometry is changed using `feature.setGeometry(new_geometry)`
ol.interaction.Snap should listen for a 'change:geometry' event on ol.Feature
More than one features can be be changed while handlingDownUpSequence
exampe: ol.interaction.Modify supports simultaneous editing of multiple features
When a geometry changes old indexed segments are searched by the feature extend and removed. New ones are added.
Old feature extent sholud be used when removing the old indexed segments.
This commit changes ol.Snap -> ol.interaction.Snap, which extends
ol.interaction.Pointer. The 'pointerdown', 'pointermove' and
'pointerup' map browser events are hanlded to edit the pixel and
coordinate properties to make them 'snap' to the closest feature.
