While deep nesting will not work on a standard `<select>`, we can
emulate it through data objects still, and just handle the display
of the data in the results.
This also means that the horrific deep-nested CSS is back to handle
the padding. I hope that will get fixed over time.
This also fixes one of the performance issues with adding array
data, as options are added at the very end instead of one by one.
This converts individual objects passed in through the `array`
parameter to `<option>` tags when Select2 is initialized. This
removes a lot of duplicate code from the `ArrayAdapter`, and keeps
everything closer to the native `<select>` element.
This introduces a breaking change from previous versions of Select2,
where the initial value for array data was blank, even if a blank
option was not present in the original array of objects. Now the
first object passed in will be selected by default, following the
behavior of a standard `<select>` element.
This breaking change does not affect `<select multiple="multiple">`
elements, which by default have no selection.
When using array data, and an option is selected, the data that is
attached to the DOM element will be run through `item` and should
have any private, automatically generated attributes added and
merged with it.
We should only be checking the option values that matter, such as
`id` and `text`, instead of checking all of the option values. This
will prevent unexpected breaking when new properties are added to
the options.
Existing properties should be covered by tests to avoid regressions.
We have to enforce ids being strings as the values of options within
a select will always be an id. This fixes an issue that we had with
array selections not being highlighted in the results.
