Panel constructor ignores dtype

[CRP]

I thought this was fixed(see issue 411), but it appears to keep resurfacing ;)

In [27]: a=Panel(items=range(2),major_axis=range(10),minor_axis=range(5),dtype=np.float32)

In [28]: a[0].dtypes
Out[28]:
0 float64
1 float64
2 float64
3 float64
4 float64

A quick debug shows that the routine form_blocks just considers generic "float", "int" etc datatypes, so maybe it does not distinguish between float32 and float64?

[adamklein]

This is an analog of another issue #622

Which currently we don't really have short-term plans to attack. In short, it converts to float64 as a "feature" to keep the API simpler

[wesm]

More work is needed in the core infrastructure to enable the use of the full NumPy dtype hierarchy without allowing the complexity to leak through to the average user

[jreback]

@wesm this should be closable for 0.11