BUG: tzinfo lost when concatenating multiindex arrays

[sinhrks]

Closes #6606.

The problem is caused by MultiIndex.append. Based on current master, MultiIndex.append works as below. The fix covers case1 and case2 which the result will be MultiIndex.

The fix is applied to concat, and also pivot_table work as expected.

import pandas as pd

idx1 = pd.Index([1.1, 1.2, 1.3])
idx2 = pd.date_range('2011-01-01', freq='D', periods=3, tz='Asia/Tokyo')
idx3 = pd.Index(['A', 'B', 'C'])

midx_lv2 = pd.MultiIndex.from_arrays([idx1, idx2])
midx_lv3 = pd.MultiIndex.from_arrays([idx1, idx2, idx3])

#1 results in MultiIndex, which nlevels is 2
midx_lv2.append(midx_lv2)

#2 results in MultiIndex, which nlevels is 2, not 3. 3rd line will be disappeared.
midx_lv2.append(midx_lv3)

#3 results in tupled Index.
midx_lv2.append(idx1)

#4 results in tupled Index.
result = midx_lv3.append(midx_lv2)

[jreback]

looks fine...pls rebase one more time

[jreback]

thanks!