Improve memory usage for computing hybrid results in the clear

[akoshelev]

This operation was OOMing on 500M inputs

memory allocation of 73551314960 bytes failed
Aborted

The reason for that was that we were keeping all the information from original reports, including AAD and match key.
The improved algorithm only keeps breakdown keys and values to perform attribution.

My back of the envelope calculation says that this improvement is enough to get us to 1 billion reports. Each entry would take ~8 bytes and for 1B entries we are looking at 8Gb memory consumption which should be doable

I ran it on 500M and it finished successfully.

[eriktaubeneck]

looks good!

[codecov]

Codecov Report

Attention: Patch coverage is 93.33333% with 3 lines in your changes missing coverage. Please review.

Project coverage is 93.28%. Comparing base (bb543f9) to head (1acf7bd).
Report is 4 commits behind head on main.

Files with missing lines	Patch %	Lines
ipa-core/src/bin/in_the_clear.rs	0.00%	2 Missing ⚠️
ipa-core/src/test_fixture/hybrid.rs	97.67%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1488   +/-   ##
=======================================
  Coverage   93.28%   93.28%           
=======================================
  Files         239      239           
  Lines       43532    43546   +14     
=======================================
+ Hits        40608    40624   +16     
+ Misses       2924     2922    -2     

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