feat: get a full prediction, not just output

[zeke]

This PR updates the template to use replicate.predictions.create instead of replicate.run, to start new users off on the right foot, with awareness that there's a whole prediction object available to them, and not just the bare output.

This is useful because:

• It gives people a prediction id, so they have a handle to associate their output with something via API or web.
• It shows stats like prediction time
• It shows the status of the prediction, and logs
• It fits better with the eventuality of teaching users to use webhooks

Working through this revealed to me that the current process for awaiting a prediction with the JS client is not so glamorous: replicate/replicate-javascript#204

Not sure I wanna merge this. Curious to hear thoughts from @replicate/hackers and @replicate/product

[deepfates]

I like this from a usability standpoint but from a wording standpoint i feel weird about showcasing replicate.wait this way. Just seems to imply that it will take a long time, while .run implies something happening right away. No strong block though

[zeke]

Changing this to a draft because it's more of a conversation piece than I thing I really want to land.