Handle crashes in multiple children

[woodruffw]

We handle multiple children correctly as of #176, but we don't handle the following case:

1. A spawns B
2. We fuzz targets within both A and B
3. B crashes
4. A crashes because it was relying on B

Currently, B's initial.dmp will be clobbered by A's initial.dmp, likely leading to nonsensical tracer results. As such, we probably need to distinguish dumps and other crash-related files by PID, e.g. initial.XXXX.dmp, crash.XXXX.json, etc. This will make the task of enumerating crash files a little more tedious, of course.

[woodruffw]

Multiple children also have consequences for index-based function targeting: each child has their own copy of call_counts in the Wizard, so our indexes are almost certainly incorrect when dealing with multiple children (this would explain the inconsistency with LibreOffice).

Might be a good reason to switch the default away from index-based targeting for the beta.