take topological ordering into account when adding/dropping dependent cols/tables

Migrated issue, originally created by lu_zero NA

Example:

roles = db.Table('roles',
    db.Column('customer_id', db.Integer, db.ForeignKey('customer.id')),
    db.Column('company_id', db.Integer, db.ForeignKey('company.id')),
#    db.Column('role_id', db.Integer, db.ForeignKey('role.id'))
)

"""
class Role(db.Model):
    id = db.Column(db.Integer, primary_key = True)
    name = db.Column(db.String(64), index = True, unique = True)
    description = db.Column(db.String(256), index = True, unique = True)
    customer = db.relationship("Customer", uselist=False, secondary=roles, backref="role")
"""

INFO [alembic.autogenerate] Detected removed table u'role'
INFO [alembic.autogenerate] Detected removed column 'roles.role_id'

sqlalchemy.exc.InternalError: (InternalError) cannot drop table role because other objects depend on it
DETAIL: constraint roles_role_id_fkey on table roles depends on table role

Inverting the order of the operation did the right thing.

Start with Alembic autogenerate's handling of detected removed tables and columns, using the roles and role example to reproduce the failure. Trace the generated operation order and verify that dependent constraints or columns are handled before their referenced tables are dropped.