Cudf: automatically infer dtypes when using read_csv

[ericmjl]

Feature request

Related to #487 (in spirit, not in technicalities), this is an issue report about reducing user surprise when interfacing with cuDF.

The pandas read_csv API does not require me to specify dtypes for my data. Even though dtype specification is an option, pandas has a sane default in which it automatically infers dtypes. To help with user adoption, this should be the default as well. As evidenced by #486, the lack of documentation reachable from a search engine was a a big stumbling block.

[BradReesWork]

I will work on getting the documentation better. The API should match Pandas where specifying the dtype is optional.

[Iroy30]

That is currently in the pipeline with PR #478

[kkraus14]

Fixed by #478