DataFrame.from_records lose dtype info on empty recarray.

[TomAugspurger]

In [2]: from pandas import *

In [3]: a = np.array([(1, 2)], dtype=[('id', 'u8'), ('value', 'i8')])

In [4]: DataFrame.from_records(a[:0]).dtypes
Out[4]:
id       float64
value    float64
dtype: object

I think the dtypes are incorrect. The record array dtypes should pass through so you get uint64 & int64.

Working on a PR now.

Split from #20802

[jbrockmendel]

closed by #40121?

[MarcoGorelli]

master shows

In [14]: a = np.array([(1, 2)], dtype=[('id', 'u8'), ('value', 'i8')])
    ...: 
    ...: DataFrame.from_records(a[:0]).dtypes
Out[14]: 
id       uint64
value     int64

could do with a test?

[jbrockmendel]

thats very nearly the test in #40121. closing