Add a comment that explains why the Classifier chain clears Variable attribues.

[grafi-tt]

This PR makes the Classifier chain better explains the importance to clear the Variable attirubtes. It prevents sloppy users defines their own chain that consumes unnecessary memory.

The docstring of chainer.links.model.Classifier says:

This is an example of chain that wraps another chain. It computes the
loss and accuracy based on a given input/label pair.

So it is reasnoable to assume that users often view and modify the source to define a custom chain. In fact, such a double memory allocation bug has actually happend in a project I'm work for.

[mitmul]

Thank you for the PR! How about putting the explanation into the docstring of the class Classifier? If you put the explanation to the docstring, both users who are modifying this class and those who will modify the code can see the explanation. So I think it's good to have it in the documentation to make it visible to all.

[grafi-tt]

@mitmul Sorry for late reply. I moved the comment to a note on docstrings (with change of wordings).

[toslunar]

LGTM

[mitmul]

Welcome back! Jenkins, test this please

[chainer-ci]

Jenkins CI test (for commit ea18ad2, target branch master) failed with status FAILURE.
(For contributors, please wait until the reviewer confirms the details of the error.)

[toslunar]

Jenkins, test this please.

[chainer-ci]

Jenkins CI test (for commit ea18ad2, target branch master) failed with status FAILURE.
(For contributors, please wait until the reviewer confirms the details of the error.)

[mitmul]

Jenkins, test this please

[chainer-ci]

Jenkins CI test (for commit ea18ad2, target branch master) succeeded!

[mitmul]

LGTM