Surprising lack of gradient clipping

I'm noticing a _surprising_ lack of gradient clipping, as it is usually a core feature.
What's the reason for its lack of existence, and how could I add it myself?

Given the wealth of functions on tensors it wouldn't be that hard to add, I'm just not sure where it would go.

Start by reading Candle's tensor APIs and tracing where training or optimization operations would accept gradient-related behavior. Clarify the expected clipping semantics and integration point with maintainers before changing code; done should include an agreed implementation and coverage for the chosen behavior.

Written by the indexing model from the issue text.