It will also be much faster to fetch the remote entities in batches
rather than one at a time through LoadEntity.
One bonus/hazard with Multi Fetch GET is that it will get deleted entities
with `visible=false`, rather than returning a HTTP Status Code 410 (Gone).
This will be the only way that we can really do proper undeletion
(Incrementing the current version by 1 is not guaranteed to work. And
if a way is moved, fetching way/full will tell us whether the childnodes
are part of the way, but not necessarily whether they exist or not.)
We must be careful never to merge deleted entities into the real graph.
e.g, a deleted node will not have a 'loc' attribute, so code that assumes
every node must have a `loc` will be broken.
So because deleted entities are very special, the output from `loadMultiple`
should only be used for conflict resolution for now.
(instead of dispatching `load` event to merge them into `history`)
This is cleaner becuase now `context` doesn't need to keep an
`altGraph` state used only for Conflict Resolution.
The conflict resolution code calls the `iD.Connection` methods directly,
and other places in the code call `loadEntity` and `loadTiles` wrappers
that merge the entities into the main history.
Because the embedded iD does not show the account
UI, Connection#userDetails was not being called, so
the commit UI didn't show the user name and image.
Fixes#1464
Connection defaults to osm.org URL and keys for the simple
case. Customization is done via connection.switch().
Externalize SourceSwitch's use of iD.data.keys.
