Use per-machine symbol_versions in policy.json

[mayeut]

The symbol versions exposed by glibc/gcc/libstdc++ are different for a single "semver" version depending on the machine running.
This commit reflects that fact by defining symbol_versions per machine.

This also aims to help automating the PEP600 transition (checking minimum common subset of symbols between distros). The fact that there were only one set of symbol versions prevented the automated check of this subset.

Things to notice:

• I couldn't get a ppc64 (big-endian) running so using the same versions as before not to introduce a regression.
• The json schema does not map to the json file but rather to the "loaded" json which keeps the same structure as before not to change the code in multiple places. One might argue it's not desirable and I can rework this in order not to change the schema while loading (but rather modify the code at appropriate places).

[codecov]

Codecov Report

Merging #262 into master will increase coverage by 0.30%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #262      +/-   ##
==========================================
+ Coverage   90.06%   90.37%   +0.30%     
==========================================
  Files          20       20              
  Lines        1057     1049       -8     
  Branches      226      224       -2     
==========================================
- Hits          952      948       -4     
+ Misses         62       60       -2     
+ Partials       43       41       -2     

Impacted Files	Coverage Δ
auditwheel/policy/__init__.py	100.00% <100.00%> (+6.66%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3fd90dd...fcf58e0. Read the comment docs.