unordered_map,unordered_set: Increase excess size

[stotko]

The excess size of unordered_map and unordered_set have been estimated based on the birthday paradox and then conservatively lowered as some amount of estimated collisions also includes entries sitting in a bucket instead of the excess list. This size adjustment has been performed to save memory. However, the underlying assumption of this estimate is a perfect uniform distribution of the hashes which is the best case scenario. With common hash functions, that distribution may be worse leading to a larger amount of required excess entries, especially at high load factors. Use the estimated number directly as it overestimates the excess list size by a factor close to 2 (about 90% collisions will typically be between exactly 2 entries). This has been tested to work better with high load factors up to 100%.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 97.46%. Comparing base (1523669) to head (f89d6d5).
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #435      +/-   ##
==========================================
+ Coverage   97.34%   97.46%   +0.11%     
==========================================
  Files          32       32              
  Lines        2524     2602      +78     
==========================================
+ Hits         2457     2536      +79     
+ Misses         67       66       -1     

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