CLN: tshift & shift could be consolidated

[max-sixty]

Particularly now that shift only works on datetime-like indexes (#11211)

They look almost the same, although not exactly:

In [11]: df=pd.DataFrame(pd.np.random.rand(5,2), index=pd.date_range(periods=5, start='2000'))

In [12]: df
Out[12]: 
                   0         1
2000-01-01  0.640148  0.781291
2000-01-02  0.261649  0.652372
2000-01-03  0.642422  0.734348
2000-01-04  0.582657  0.601868
2000-01-05  0.848645  0.078437

In [13]: df.shift()
Out[13]: 
                   0         1
2000-01-01       NaN       NaN
2000-01-02  0.640148  0.781291
2000-01-03  0.261649  0.652372
2000-01-04  0.642422  0.734348
2000-01-05  0.582657  0.601868

In [14]: df.tshift()
Out[14]: 
                   0         1
2000-01-02  0.640148  0.781291
2000-01-03  0.261649  0.652372
2000-01-04  0.642422  0.734348
2000-01-05  0.582657  0.601868
2000-01-06  0.848645  0.078437

In [15]: df.shift(freq='D')
Out[15]: 
                   0         1
2000-01-02  0.640148  0.781291
2000-01-03  0.261649  0.652372
2000-01-04  0.642422  0.734348
2000-01-05  0.582657  0.601868
2000-01-06  0.848645  0.078437

In [16]: df.tshift(freq='D')
Out[16]: 
                   0         1
2000-01-02  0.640148  0.781291
2000-01-03  0.261649  0.652372
2000-01-04  0.642422  0.734348
2000-01-05  0.582657  0.601868
2000-01-06  0.848645  0.078437

[jreback]

yes, shift is implemented in terms of tshift (which is simply missing the axis arg).

So would take a PR to deprecate .tshift, move impl to .shift and change tests/such.

[IamGianluca]

I'm having a look at this issue. I'll send a PR hopefully by end of next week.