fix(share/eds): remove bitswap cache wrap from eds store

[walldiss]

Two layers of caching were implemented for bitswap requests in full nodes using EDS store:

A bitswap cache that caches individual shares.
An internal EDS store cache that caches accessors.
In the event of a cache miss, both the internal cache of the EDS store and the bitswap cache will be populated with the same share, introducing redundancy in caching. Moreover, having a per-share cache doesn't seem to make much sense for full nodes, given the astronomically small chances of requesting the same share from a block that is not cached in the EDS cache.

The pull request removes the per-share cache as it brings close to no value while consuming RAM. Your understanding and consideration of this change are greatly appreciated.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 0% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 44.83%. Comparing base (2469e7a) to head (1778be5).
Report is 15 commits behind head on main.

Files	Patch %	Lines
nodebuilder/p2p/bitswap.go	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3101      +/-   ##
==========================================
- Coverage   44.83%   44.83%   -0.01%     
==========================================
  Files         265      263       -2     
  Lines       14620    14628       +8     
==========================================
+ Hits         6555     6558       +3     
- Misses       7313     7321       +8     
+ Partials      752      749       -3     

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

[Wondertan]

Iirc, it doesnt cache the full blocks there but their local availability for Has method via bloomfilters. It is an optimization for bitswap to avoid doing Hazes(reads). This might still be helpful, but needs more thoughts.

[walldiss]

The blockstore cache implemented in our full node does not update when a block is acquired from shrex. If a block request fails because it's absent in the eds-store, the cache erroneously flags it as 'not available'. Consequently, when the block is later synchronized, the cache fails to reflect this update. This discrepancy leads to the risk of generating false 'not available' statuses:

1. A block not stored receives an initial request (1), resulting in a 'not available' cache entry.
2. The block is subsequently retrieved and stored.
3. A second request (2) for the same block consults the cache, mistakenly producing a 'not available' response.

Considering this, disabling the blockstore cache might be advisable to prevent such inaccuracies.

[walldiss]

This is not a problem for Light nodes, as blockstore cache will be updated on successful sampling.