Don't rely on ResultProxy.keys for column names

[dhirschfeld]

These names can be truncated/translated so won't always map back to the
column names

[dhirschfeld]

import sqlalchemy as sa

col_name = 'a_column_which_has_more_than_24_chars'
table = sa.Table(
    'Test', sa.MetaData(),
    sa.Column('id', sa.Integer, primary_key=True),
    sa.Column(col_name, sa.Float(32), nullable=False),
)

In sqlalchemy a compiled expression can include truncated column names

In [88]: from sqlalchemy.dialects.oracle.cx_oracle import OracleDialect_cx_oracle

In [89]: dialect = OracleDialect_cx_oracle()

In [90]: s = table.select().limit(10)

In [91]: print(s)
SELECT "Test".id, "Test".a_column_which_has_more_than_24_chars 
FROM "Test"
 LIMIT :param_1

In [92]: print(s.compile(dialect=dialect))
SELECT id, a_column_which_has_more__1 
FROM (SELECT "Test".id AS id, "Test".a_column_which_has_more_than_24_chars AS a_column_which_has_more__1 
FROM "Test") 
WHERE ROWNUM <= :param_1

Currently odo relies on ResultProxy.keys to get the column names without any code to handle potentially truncated names which can result in errors like below:

In [41]: odo(data.head(), pd.DataFrame)
Traceback (most recent call last):

  File "<ipython-input-41-15ecd5ef9d0d>", line 1, in <module>
    odo(data.head(), pd.DataFrame)

  File "C:\Miniconda3\lib\site-packages\odo\odo.py", line 91, in odo
    return into(target, source, **kwargs)

  File "C:\Miniconda3\lib\site-packages\multipledispatch\dispatcher.py", line 164, in __call__
    return func(*args, **kwargs)

  File "C:\Miniconda3\lib\site-packages\blaze\compute\core.py", line 379, in into
    return into(a, result, **kwargs)

  File "C:\Miniconda3\lib\site-packages\multipledispatch\dispatcher.py", line 164, in __call__
    return func(*args, **kwargs)

  File "C:\Miniconda3\lib\site-packages\odo\into.py", line 43, in wrapped
    return f(*args, **kwargs)

  File "C:\Miniconda3\lib\site-packages\odo\into.py", line 53, in into_type
    return convert(a, b, dshape=dshape, **kwargs)

  File "C:\Miniconda3\lib\site-packages\odo\core.py", line 83, in __call__
    return _transform(self.graph, *args, **kwargs)

  File "C:\Miniconda3\lib\site-packages\odo\core.py", line 106, in _transform
    x = f(x, excluded_edges=excluded_edges, **kwargs)

  File "C:\Miniconda3\lib\site-packages\odo\backends\sql.py", line 756, in select_or_selectable_to_frame
    dtype=[(str(c), dtypes[c]) for c in columns]))

  File "C:\Miniconda3\lib\site-packages\odo\backends\sql.py", line 756, in <listcomp>
    dtype=[(str(c), dtypes[c]) for c in columns]))

KeyError: 'pasaavailability_schedul_1'

This can be avoided entirely by instead getting the name from each column in the Select.columns collection.

[dhirschfeld]

Test failure is the same random failure which is fixed by #557

[dhirschfeld]

Would also be great to get this one merged

[llllllllll]

looks good, thanks!