fix(share/ipld): Remove proofs from leaked ctx

[walldiss]

Inside tee-getter ProofAdder is added to ctx to collect proofs and allow eds store to read them, without extra recompute. Ctx provided to store.Put method, will be used for register shard, that in some rare cases takes long time and returns, before result returned from operation. The provided context is still leaked to async operation, and it prevents golang GC from collecting memory allocated for ProofAdder, because there is still possibility for code to reach it.

Solution
Dagstore register shard operation does not need to access to proofs, so the associated memory could be reclaimed right after proofs are extracted from ctx. For convenience it will happen after put operation

Solves first point of #2582

[codecov-commenter]

Codecov Report

Merging #2574 (a0d4a7e) into main (003c2c4) will increase coverage by 0.05%.
Report is 1 commits behind head on main.
The diff coverage is 43.47%.

@@            Coverage Diff             @@
##             main    #2574      +/-   ##
==========================================
+ Coverage   51.01%   51.07%   +0.05%     
==========================================
  Files         158      158              
  Lines       10398    10428      +30     
==========================================
+ Hits         5305     5326      +21     
- Misses       4627     4636       +9     
  Partials      466      466              

Files Changed	Coverage Δ
share/ipld/nmt_adder.go	45.71% <0.00%> (-3.78%)	⬇️
share/eds/eds.go	63.73% <40.00%> (-0.16%)	⬇️
core/exchange.go	35.84% <50.00%> (+0.55%)	⬆️
core/listener.go	53.50% <100.00%> (-1.85%)	⬇️
share/getters/tee.go	42.50% <100.00%> (+4.66%)	⬆️

... and 4 files with indirect coverage changes

[Wondertan]

I assume you've tested this and seen stable RAM usage on Graphana

[Wondertan]

The fact that proofs are leaking is a hint that the context is leaking. Because if the context is cleared up, the proofs would be cleared up as well. WDYT?

[walldiss]

The fact that proofs are leaking is a hint that the context is leaking. Because if the context is cleared up, the proofs would be cleared up as well. WDYT?

I agree, Thats why PR is named for issue of leaked ctx, but not just proofs "Remove proofs from leaked ctx". Provided solution is a hotfix for latest release to prevent leak. Proper investigation of leaked ctx will take some time, since it is passed into many complex components, like ipld, bitswap, dagstore, badger. I'll create an issue to investigate where the ctx is leaked.

#2582