fix(dwarf): Resolve symbol names by exact address

[loewenheim]

This changes the lookup of functions in the symbol table so that the symbol must have the same starting address as the function, rather than merely covering the entire function.

[codecov-commenter]

Codecov Report

Merging #510 (ab86689) into master (7810fee) will decrease coverage by 0.07%.
The diff coverage is 100.00%.

❗ Current head ab86689 differs from pull request most recent head 393b5f1. Consider uploading reports for the commit 393b5f1 to get more accurate results

@@            Coverage Diff             @@
##           master     #510      +/-   ##
==========================================
- Coverage   68.73%   68.66%   -0.08%     
==========================================
  Files          84       84              
  Lines       17500    17503       +3     
==========================================
- Hits        12029    12018      -11     
- Misses       5471     5485      +14     

[jan-auer]

cc @mitsuhiko.

We found a case where we incorrectly matched the last (unbounded) symbol in the symbol table, but the start address was far off the DWARF function record. This made me question the approach, and I think this exact lookup is actually correct.

Do you remember which compilers were generating inferior DWARF name attributes? I'm looking for a regression test.

[mitsuhiko]

Sorry for the late reply but I do not remember. There is also a non zero chance that we had to do this due to another bug somewhere else. That said, the easiest way would be to sample this on production data now.