Handle JSON generation like other errors instead of bailing out

[GuillaumeGomez]

I tested one of the failed crate on docs.rs and couldn't reproduce the bug locally, so I assume the build could be retried and therefore this error should be handled like the others.

[syphar]

my original reasoning was that the json build might crash and I wouldn't want to have the html build failing when we only have a rustdoc-json bug/crash.

Or do you think it's stable enough now?

I tested one of the failed crate on docs.rs and couldn't reproduce the bug locally, so I assume the build could be retried and therefore this error should be handled like the others.

Was there a different nightly version for the initial build and the re-run? Or the same and just re-running was what's needed?

View changes since this review

[syphar]

Also the tests fail for proc-macro crates, remembering now that there was an edge case about json&proc-macros I don't remember?

[GuillaumeGomez]

Let's keep things as is then. I saw an issue in sentry listing this but maybe I misunderstood the problem. Anyway, closing. :)