fix(anyhow): attach stacktrace only if anyhow error has backtrace

[lcian]

This is similar to #755 but for the anyhow integration

Without this change, if we're running without RUST_BACKTRACE set, then backtrace will have value "<disabled>".
We would try to parse it and get an empty vec as a result.
The consequence is that all anyhows would be grouped in a single issue due to having the same (empty) stack trace.

After the change, there will be no stack trace, so grouping will happen according to the error message.

Side effect: when running with attach_stacktrace: true, sentry-backtrace will kick in and provide a backtrace (this didn't happen before because we had a stack trace, even though it was an empty one).
It could be argued whether we want this or not.

Closes #745

[Swatinem]

nice! thanks for fixing this and providing a good explanation.

Maybe it might be worth still checking exc.stacktrace if its empty, and clearing it in that case?

[lcian]

Sorry the stacktrace is not empty, it's going to have a single frame which will contain either disabled or unsupported.
We wrongly parse that and believe it's a frame.
This is due to the fact that anyhow uses some custom logic to format the backtrace https://github.com/dtolnay/anyhow/blob/master/src/backtrace.rs. Luckily they provide backtrace.status() to check it.
In general we already return None in Stacktrace::from_frames_reversed if the vec of frames is empty, so I don't think we need to check twice.