Symbol resolution for global variables (not functions)

This actually used to work with the old `ProcessSymbols` code, and I think we have regressed when moving to the C libelf-based implementation in libbcc. It would still be beneficial to support global variables, e.g. to resolve global mutexes for the lockstat tool.

Start by comparing the old ProcessSymbols implementation with the current C libelf-based symbol resolution in libbcc. Trace how the lockstat tool requests symbols, then verify that global variables, including global mutexes, resolve successfully without regressing function resolution.

Written by the indexing model from the issue text.