Change the memory queue implementation to not pre-allocate capacity objects.

[bogdandrutu]

Change the sized channel to use a "fifo" queue without pre-allocated capacity:

• Allows us to soon allow different sizers like bytes, which otherwise would pre-allocate way too much memory. E.g. if configure with a limit of 1GB would pre-allocate 16GB for the buffered channel only.
• Does not pre-allocate memory when not needed.

Before:

cpu: Apple M2 Max
BenchmarkOffer
BenchmarkOffer/requests_based
BenchmarkOffer/requests_based-12         	 5559658	       207.3 ns/op	       0 B/op	       0 allocs/op
BenchmarkOffer/items_based
BenchmarkOffer/items_based-12            	 6025567	       209.4 ns/op	       0 B/op	       0 allocs/op

After:

cpu: Apple M2 Max
BenchmarkOffer
BenchmarkOffer/requests_based
BenchmarkOffer/requests_based-12         	 4904857	       224.3 ns/op	      48 B/op	       1 allocs/op
BenchmarkOffer/items_based
BenchmarkOffer/items_based-12            	 5470143	       229.9 ns/op	      48 B/op	       1 allocs/op

Overall there is a ~20ns performance downgrade but this is expected since no memory pre-allocation.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 91.68%. Comparing base (e6df9d8) to head (2144722).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #12070      +/-   ##
==========================================
+ Coverage   91.66%   91.68%   +0.01%     
==========================================
  Files         455      455              
  Lines       24005    24039      +34     
==========================================
+ Hits        22005    22039      +34     
  Misses       1629     1629              
  Partials      371      371              

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

[dmitryax]

Nice!