transformations: stencil bufferization polish

[PapyChacal]

Step back on the idea of allocating all apply results to then fold away and instead just fold stores into apply. It's cheaper, simpler, and so far works just as well!

Also rely on the new side-effect value information to improve side-effect analysis.

Both changes allow to bufferize a full open earth compiler example now!

Minor test changes, the bufferization used to blindly allocate buffers for unused results, now it scraps them. I guess it's debatable if that should be its role? Note that it does so by reusing the canonicalization pattern, no logic is actually duplicated.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.88%. Comparing base (7473b87) to head (a914a27).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3001      +/-   ##
==========================================
+ Coverage   89.86%   89.88%   +0.02%     
==========================================
  Files         415      415              
  Lines       52078    52095      +17     
  Branches     8048     8047       -1     
==========================================
+ Hits        46799    46826      +27     
+ Misses       3984     3977       -7     
+ Partials     1295     1292       -3     

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