PERF: Series.dropna with non-nan dtype blocks

[sinhrks]

Minor improvement when dropna actually does nothing. If OK for 0.17, I'll add a release note.

import numpy as np
import pandas as pd

s = pd.Series(np.random.randint(0, 1000, 50000000))

# before
%timeit s.dropna()
#1 loops, best of 3: 1.58 s per loop

# after
%timeit s.dropna()
#1 loops, best of 3: 568 ms per loop

[jorisvandenbossche]

the function that actually get times should start with 'time'

[kawochen]

Also this dropna only has effect for each time setup is run, which is not every iteration
And set seed (not sure)?

[jreback]

sure, pls add a note in 0.17.1 whatsnew

[jreback]

@sinhrks looks good, can you rebase / add a release note

[sinhrks]

Rebased, and added release note. Followings are asv results added in this PR.

    before     after       ratio
  [a89b96d4] [b2cb56a8]
    10.76ms    10.60ms      0.99  series_methods.series_dropna_datetime.time_series_dropna_datetime
-   12.07ms   726.61μs      0.06  series_methods.series_dropna_int64.time_series_dropna_int64

[jreback]

thanks!