improve drsym_unix_lookup_symbol linear walk performance
On a large app, drcachesim's -record_heap can take a long time (1.5 minutes!) to query just the handful of heap functions it's looking for. (We recently added operator new and that rose to 2:40).
It looks like drsyms does a linear walk for symbol lookup! We should have it build a hashtable or sthg, or if we're worried about space take in a list of things to look for (or trace_func.cpp should change to enumerate w/ its own callback to do a single walk).
Even for addr lookup, for ELF it looks linear. MachO and PECOFF have a binary search.
Or, add a flag --dynsym_only where func_trace never calls into drsyms, as a drcachesim-focused fix.