Improve debug information handling

[javierhonduco]

Right now we try to get ahold of debug information as soon as possible, to increase the chances of being able to open them before the get deleted / unmounted. The current implementation however has several shortcomings:

• too many object files are opened for a very long period of time, and in some cases there might not even be debug information in them;
• the current implementation doesn't account different debug information storage implementation;

Ideally an improved implementation would reduce the amount of file descriptors open, while still keeping the chance of successfully being able to deal with the debug information as high as possible.

Note: right now the current "potential" container path to absolute root mount view, is a bit wasteful in terms of memory. We should only prepend /proc/<pid>/root if the process is running in another mount namespace.

[javierhonduco]

While it's not a comprehensive fix, 5b9b2b2 improves the chances of correct symbolisation for workloads that aren't containerised