Performance improves by ~7% by optimizing allocation of vectors and strings
iin module_symbol_index and module_symbol_index_node before doing the indexing.

Performance improves by ~15% by optimizing the allocation of the "seen_attrs"
vector in the DwarfDieDecoder.

This was measured by running the normall-commented-out test
ModuleSymbolIndex.BenchmarkIndexing and giving it an ELF binary of a fully
symbolized release build of Chrome.

There should be no functional change.

Change-Id: I107a3d651ac05e9b2959da88a06e463173eae199