fix: correct __all__ names in classification/loss modules

[shivamlalakiya]

What

conftr.py, confts.py and scpo.py each declare an __all__ that names a
class that does not exist in the module:

classification/loss/conftr.py:8   __all__ = ["ConfTr"]   # class defined is ConfTrLoss
classification/loss/confts.py:8   __all__ = ["ConfTS"]   # class defined is ConfTSLoss
classification/loss/scpo.py:8     __all__ = ["SCPO"]     # class defined is SCPOLoss

A wildcard import from any of the three modules directly raises
AttributeError instead of binding the class. Reproduced against the
1.2.1 wheel from PyPI:

>>> from torchcp.classification.loss.conftr import *
AttributeError: module 'torchcp.classification.loss.conftr' has no attribute 'ConfTr'
>>> from torchcp.classification.loss.confts import *
AttributeError: module 'torchcp.classification.loss.confts' has no attribute 'ConfTS'
>>> from torchcp.classification.loss.scpo import *
AttributeError: module 'torchcp.classification.loss.scpo' has no attribute 'SCPO'

The named-import path is unaffected, because torchcp/classification/loss/__init__.py
imports the real class names directly rather than going through __all__:

>>> from torchcp.classification.loss import ConfTrLoss, ConfTSLoss, SCPOLoss
# ok

So this only bites a caller who does from torchcp.classification.loss.conftr import *
(or the confts / scpo equivalents) instead of importing from the package.

Fix

One line changed per file: __all__ now names the class that is actually
defined (ConfTrLoss, ConfTSLoss, SCPOLoss). Verified the wildcard import
resolves correctly against the patched files:

>>> from torchcp.classification.loss.conftr import *; ConfTrLoss
<class 'torchcp.classification.loss.conftr.ConfTrLoss'>
>>> from torchcp.classification.loss.confts import *; ConfTSLoss
<class 'torchcp.classification.loss.confts.ConfTSLoss'>
>>> from torchcp.classification.loss.scpo import *; SCPOLoss
<class 'torchcp.classification.loss.scpo.SCPOLoss'>

This is a pure one-liner per file, no other behaviour changes.