Now users can supply data to the attribute "data-select2-tags" to provide
existing tags.

Multiple Primary keys for sqla backend

bugfix: inline form 'new' template - now submits all added fields correc...

Fix "NoneType" has no attr 'get' on subdocumments

in this line  https://github.com/mrjoes/flask-admin/blob/master/flask_admin/contrib/mongoengine/form.py#L44
None causes "NoneType" has no attr 'get'

`contrib.sqla.view.ModelView.action_delete()` now is able to mass-delete models with multiple primary keys. Therefore, two new helper functions has been introduced in `contribg.sqla.tools`:

* `tuple_operator_in()` that fakes the `sqlalchemy.tuple_` operator for engines, that do not support the `tuple_` operator. (like sqlite)
* `get_query_for_ids()` that returns a query, object that contains all objects identified by the `ids`.

Both functions have docstrings with more information.

The logic of the `action_delete()` did not change, as for objects with a single primary key, the codepath did not change at all.

* `contrib.sqla.tools.get_primary_key()` now works with inhteritance *and* multiple primary keys.

Every primary key is checked if it is a inherited one, and if it's inherited, the *current* column.key is returned. I *think* it should be safe just to use *prop.expression* here, but I am not absolutely sure.

( see http://docs.sqlalchemy.org/en/rel_0_8/orm/internals.html?highlight=columnproperty#sqlalchemy.orm.properties.ColumnProperty.expression )

* New helper-function `contrib.sqla.tools.has_multiple_pks(model)`
* The unique validator is only added to the primary key column in `converter()`, when the model does not have multiple primary keys.
* Removed option `form_excluded_pk_columns_from_unique_validation`
* documentation updated
* example updated

Flask-Admin can now handle multiple primary keys to some extent. `get_primary_key()` and `get_pk_value()` where the simple, but important changes. When building the url in `list.html` via `url_for` a url with multiple `&id=x&id=y` is built. As the order as preserved and the request args are from type `werkzeug.datastructures.MultiDict`, the multiple PKs can be grabbed in the correct order an push into `get_one()` which actually has no problems with multiple keys, as sqlalchemy's `query.get()` works exactely in this way.

* `contrib.sqla.tools.get_primary_key()` returns a tuple of all primary keys, if there are more than one PKs
* `contrib.sqla.view.get_pk_value()` returns a tuple with the values of all primary keys, if there are more than one PKs
* `model.helper.get_mdict_item_or_list()` return the single item of a given MultiDict for the given key or the whole list of items, if there are more than one.
* `model.base.BaseModelView.[edit_view|delete_view]` can work with a single *id* or multiple ones, by using `get_mdict_item_or_list()`
* New option `form_excluded_pk_columns_from_unique_validation` for `model.base.BaseModelView` including docstring.
* Option is evaluated in `contrib.sqla.form.AdminModelConverter.()`

Tests and documentation is included. An example file showing the feature is in `sqla/examples/multiplepk.py`

The changes itself are very small, although it took me hours to go there - the MultiDict saved me! I will change my app I am currently working on to multiple primary keys this week and will push updates if necessary.

Fix example in column_type_formatters docstring

process inherited primary keys correctly

Two new functions on `contrib.sql.tools`:

  - `is_inherited_primary_key(prop)` : Returns True if the `ColumnProperty` **prop** is an inherited primary key. Occuring Exceptions are not handled but raised. *Should* work with arbitrary levels of inheritance.
  - `get_column_for_current_model(prop)` : Return the `Column` that belongs to the model of the `ColumnProperty` **prop**. So if the ColumnProperty is for the parent, the Column for the parent ist returned. If the ColumnProperty if for the Child, the Column for the child is returned. Works with arbitrary levels of inheritance.

`AdminModelConverter.convert()`  and `ModelView.scaffold_list_columns()` are adapted to use the new functions.

When using joined table inheritance (http://docs.sqlalchemy.org/en/latest/orm/inheritance.html), it is common practice to name the pk-property the same like the pk-property of the parent. The child-property ist a pk itself and has a foreign-key relationship to the pk-property of the parent.

Example:

    class BaseWahl(Model):
        __tablename__ = 'basewahlen'

        id = Column(Integer, primary_key=True, autoincrement=True)
        discriminator = db.Column(db.String(50))

        __mapper_args__ = {
        'polymorphic_identity':'basewahl',
        'polymorphic_on': discriminator
        }

    class Wahl(BaseWahl):
        __tablename__ = 'wahlen'

        id = Column(Integer, ForeignKey('basewahlen.id'), primary_key=True)
        __mapper_args__ = {
            'polymorphic_identity':'wahl',
        }

`AdminModelConverter.convert()` does not allow Properties with multiple columns, but will raise a `TypeError`. I changed it into the following way:

- If
  - more than 1 column for the property
  - all columns are primary keys
  - only *one* does not have a Foreign key
  - only one column corresponds to the current model
- select the column, that corresponds to the current model

I applied the same code to `ModelView.scaffold_list_columns()` and extended it, the primary key in this special constellation actually is *not* ignored, even if it has a foreign key property, so it can be displayed using `column_display_pk = True`.

This solution actually works for me. I do not have enough insight into *Flask-Admin* and definitely even less into *sqlalchemy* to think my solution is the correct one for every situation, but it may be something to think about.