Fix TimedeltaIndex constructor fastpath when name is set

[shoyer]

Fixes #10025

[shoyer]

For the record, I don't think it is actually a good idea to return an existing object from an index constructor. This leads to weird things like:

new_idx = pd.Index(old_idx)
new_index.name = 'foo'
# also sets old_idx.name!

We should preserve the old data and hashtable, if possible, but not everything on an Index object is immutable (name/names is the main exception).

[jreback]

you are setting the name directly, use set_names to get a new object

In [24]: i = Index(range(5),name='foo')

In [25]: i2 = Index(i,name='bar')

In [26]: i3 = i2.set_names('baz')

In [27]: i.name
Out[27]: 'foo'

In [28]: i2.name
Out[28]: 'bar'

In [29]: i3.name
Out[29]: 'baz'

[shoyer]

@jreback do you really think it is a good idea for the Index constructor to return the same object that was passed in? I agree that set_names is cleaner here...

[jreback]

@shoyer its prob IS a bug that it is doing that. I think that it should be a shallow copy always. See what the other indexes do. Certainly no tests for this type of aliasing I'll bet.

[jreback]

thanks!