potential mismatch of data type for csv column

Due to the way of discovering the datashape of a csv file:

def discover_csv(c, nrows=1000, **kwargs):

by default, it only checks the first 1000 rows, so it is possible (For example, in a column with sparse data -- mostly 0 with scattered float number) that later rows may contain data of a different type (which does not appear in the 1000 sampled rows) and cause dtype error when actually loading the data.

More over, even when the user pass in a dshape parameter, it is not taken by the CSV constructor.