Disable cache when downloading file

[ashleytqy]

Fixes #3251

The route handler (NbconvertFileHandler) that processes a download request did not have cache-control header so this adds the header so ensure that the downloaded pdf / markdown etc. file will always be the most recent (not cached).

[takluyver]

Was it actually computing the etags wrong, or was it the Cache-Control change that fixed it? Looking through tornado's code, it looks like it computes the etag by hashing the content when we call self.finish(), which I think should work.

[takluyver]

(I'd assumed before that tornado wasn't setting etags by default except for static files, and that we could try setting it from our code)

[ashleytqy]

I think the Cache-Control fixed it – I'll remove the compute_etag method!

[ashleytqy]

hi @takluyver ! does this look good to you?

[takluyver]

Oh, sorry, it dropped off my radar. It does look good, thanks. I assume you've tested that it fixes the issue?

I think we might also need the same in the respond_zip() function; that's used when nbconvert produces multiple files, so we assemble them into a zip file to be downloaded.

[ashleytqy]

@takluyver yup! i’ll also add the change to respond_zip() !

[takluyver]

Thanks :-)