[PyTorch] hotfix r0.17.2

[AnirudhDagar]

Description of changes:
The following sections are fixed:

• The hotdog class in imagenet is number 934 instead of 713.
• 1/2 is added back in mseloss for chapter_optimization in train_concise_ch11 because the plots won't show since without the 1/2 the loss gets doubled and the ylim of the training curve has been fixed between 0.22 and 0.35. Without the 1/2 factor the loss never shows up on the plots.
• lr=0.01 makes learning smoother in the updated notebooks.
• l.mean() instead of l.sum() (in the weight-decay section) is more stable, producing the expected plots.

By submitting this pull request, I confirm that you can use, modify,
copy, and redistribute this contribution, under the terms of your
choice.

[mli]

Job d2l-en/PR-2009/1 is complete.
Check the results at http://preview.d2l.ai/d2l-en/PR-2009/

[astonzhang]

Thanks. I'll move the comments # Note: MSELoss computes squared error without the 1/2 factor closer to the multiplication by 1/2 to improve readability.