Conv Transpose 1d not Implemented for CUDA.

Hey!

Just trying to reimplement something in Candle from the ground up for inference (no need for training). It's a fairly chunky model but I'm at the last hurdle.

So first of all thank you for all the work done so far. I think despite the docs not being fleshed out yet, it has actually been fairly intuitive to get around the crate for my purposes - so kudos to that!

I've hit a wall just now with the `conv-transpose1d` being not implemented on a CUDA backend. I saw that this PR added some boilerplate for it (https://github.com/huggingface/candle/pull/1251).

I am wondering how difficult would it be to populate it or if there are any work arounds right now for a CUDA backend?

I have had a look at the crate a tiny bit and would love to do a PR for it, but I don't really know where to start, both in terms of Candle, and actually what the OP should do in practice, so I will go off now and have a look/read, but any advice on how to proceed is really appreciated.

Start by reviewing PR #1251 and tracing the conv-transpose1d entry point into Candle’s CUDA backend. Compare the available backend implementations and identify the expected behavior for CUDA. Done means conv-transpose1d no longer reports as unimplemented and works for the relevant inference path, with coverage added where the project’s existing tests allow.

Written by the indexing model from the issue text.