Better way of specifying column names in CSV objects

Many character separated files come without column names. We often want to specify column names while letting blaze, datashape, and the appropriate backend perform type inference. One _can_ do this while constructing `Data` objects, but you have to pass in the entire datashape which for anything more than a few columns is extremely annoying.

What I'd like is something like this:

``` python
CSV('file.csv', fields=[...])
resource('file.csv', fields=[...])
Data('file.csv', fields=[...])
```

There are a few implementation details to juggle here.
- [ ] should `fields` be for labeling only or should it also be for subsetting similar to the `usecols` argument in `pandas.read_csv`? I vote labeling only.
- [ ] `fields` should definitely take precedence over field names in the file.
- [ ] We should see if it's possible to validate when reading that `len(csv.fields) == `