Speedup FrametoArray serializer for ChunkStore

[BaiBaiHi]

Summary
Speedup FrametoArray serializer for ChunkStore by removing intermediate DataFrame construction.
The majority of the time spent for serialization is spent on the following:

1. Constructing intermediate DataFrames
2. Setting the index:
   
   

By removing the intermediate DataFrame construction (so we only use numpy arrays and construct the DataFrame at the very end) and constructing the index separately, we can speed the serialization up significantly.

Performance comparisons

No Index - Series

    df = pd.Series(range(100))
    a = FrametoArraySerializer().serialize(df)

1. Single Chunk:
   
2. Multiple Chunks (data in list):
   

With Index - Series

    df = pd.Series(range(100), index=pd.Index(range(100), name='A'))
    a = FrametoArraySerializer().serialize(df)

1. Single Chunk:
   
2. Multiple Chunks (data in list):
   

No Index - Multiple columns

    df = pd.DataFrame(np.random.randint(0, 100, size=(100, 4)),
                      columns=list('ABCD'))
    a = FrametoArraySerializer().serialize(df)

1. Single chunk (shape: (100, 4))
   

2. Multiple chunks (data in list):
   

With Index - Multiple columns

    df = pd.DataFrame(np.random.randint(0, 100, size=(100, 4)),
                      columns=list('ABCD'))
    df = df.set_index(['A'])
    a = FrametoArraySerializer().serialize(df)

1. Single chunk (shape: (100, 4))
   

2. Multiple chunks (data in list):
   

[TomTaylorLondon]

LGTM