BUG: SparseDataFrame may not preserve passed dtype

[sinhrks]

• tests added / passed
• passes git diff upstream/master | flake8 --diff
• whatsnew entry

SparseDataFrame may reset passed sparse data's dtype and fill_value as below.

on current master:

arr = pd.SparseArray([1, 0, 3, 0], dtype=np.int64, fill_value=0)

# OK, dtype and fill_value are kept
s = pd.SparseSeries(arr, name='x')
s.dtype, s.fill_value
(dtype('int64'), 0)

# NG, dtype is reset
df = pd.SparseDataFrame(s)
df['x'].dtype, df['x'].fill_value
(dtype('float64'), 0)

# NG, dtype is reset
df = pd.SparseDataFrame({'x': s})
df['x'].dtype, df['x'].fill_value
(dtype('float64'), 0)

# NG, dtype and fill_value are reset
df = pd.SparseDataFrame({'x': arr})
df['x'].dtype, df['x'].fill_value
(dtype('float64'), nan)

[codecov-io]

Current coverage is 85.28% (diff: 100%)

Merging #13866 into master will decrease coverage by <.01%

@@             master     #13866   diff @@
==========================================
  Files           139        139          
  Lines         50011      50008     -3   
  Methods           0          0          
  Messages          0          0          
  Branches          0          0          
==========================================
- Hits          42650      42647     -3   
  Misses         7361       7361          
  Partials          0          0          

Powered by Codecov. Last update 768bf49...dea69ed

[jreback]

thanks!