Fix potential panic when reporting nom errors

[danielocfb]

We have gotten reports of a crash caused by numeric overflow when attempting to parse a Breakpad file. The reason lies in the error conversion functionality that nom provides. Specifically, the nom::error::convert_error() function expects errors with a context that derefs to an str slice. In order to fulfill that contract we convert everything to Cow.
However, the function also implicitly assumes that substrings belong to the same input slice. That is not the case, and cannot be easily fixed, because there is no sane way of "relocating" arbitrary byte slice after a lossy string conversion. The requirement of derefing into a str while also mapping to byte slices and are assumed to be subslices of each other just seems plain broken.
This change imports a copy of the convert_error() function and fixes up the broken bits.
The upstream issue touching on this problem is:
rust-bakery/nom#1696

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 94.47%. Comparing base (f371817) to head (966a7cf).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #658      +/-   ##
==========================================
+ Coverage   94.39%   94.47%   +0.07%     
==========================================
  Files          47       47              
  Lines        9159     9291     +132     
==========================================
+ Hits         8646     8778     +132     
  Misses        513      513              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.