Should same-type conversions with a different dshape cast their input to conform to the specified dshape?

Sometimes we may want to change the types of some of the columns in a table or cast an `int64` array to a float array. Should `odo` do this?

For example:

``` python
import pandas as pd
df = pd.DataFrame(dict(ints=list(range(5))))
newdf = odo(df, pd.DataFrame, dshape='var * {ints: float64}')
```

We already do this when going from type `a` to type `b`, but not when going from type `a` to type `a`.

cc @llllllllll