Respect original filename when cached example files are downloaded

[freddyaboulton]

Description

Fixes #2140

The downloaded filename is the same as what is returned from the prediction function now but you'll note that the extension for the file in the Model3D demo is messed up.

This is because when the output is cached, we return a base64 representation of the file as opposed to serving the file itself. The data streams for Model3D files have application/x-tgif and application/octet-stream mime types and I think Chrome can't guess the file extension for those mime types.

To get around this we'd have to serve the file for examples too but decided to hold off and hear other's thoughts on the matter.

Checklist:

• I have performed a self-review of my own code
• My code follows the style guidelines of this project
• I have commented my code in hard-to-understand areas
• I have made corresponding changes to the documentation
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes

[github-actions]

All the demos for this PR have been deployed at https://huggingface.co/spaces/gradio-pr-deploys/pr-2145-all-demos

[abidlabs]

Just merged 3.2 in here so the tests pass

[abidlabs]

LGTM, but there is a deeper problem that you point out, which is the for cached file examples, we first serialize them to base64 (thereby losing the extension information), and then the browser has to convert them to a file for download. We really should avoid this, but for now, let's file an issue and take a look later @freddyaboulton?

[freddyaboulton]

@abidlabs Thanks for the review! Filed #2187