read_csv treats both inf and -inf as very large negative number

[lodagro]

from mailing list, same discussion led to #2026

In [42]: data = """\
   ....: ,A
   ....: a,inf
   ....: b,-inf
   ....: """

In [43]: df = pd.read_csv(StringIO(data), index_col=0)

In [44]: df
Out[44]: 
                     A
a -9223372036854775808
b -9223372036854775808

In [45]: df.dtypes
Out[45]: A    int64

[aflaxman]

Here is a test and a patch for this bug: aflaxman@d11eea2

I'm planning to make a pull request which combines it with work on #2026 and #1919.

[wesm]

I added a workaround the inf/-inf parsing issue with something with a bit less overhead (fewer string comparisons). I'll address the other PR's separately

[aflaxman]

thanks!