Remove async cache provider

[javitonino]

Description

• Removes async cache, unified into a single implementation of cache (sync).
• Copied safe-deletion logic from async to sync cache.
• Changed grpc servers and replication to use the sync cache, obtaining the shards from inside a blocking task (spawn_blocking).

How was this PR tested?

Tests and basic tinkering. Will probably require some performance testing in stage.

[shortcut-integration]

This pull request has been linked to Shortcut Story #8277: Simplify shard cache layer.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (5483d32) 82.16% compared to head (b0cc329) 82.21%.
Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1720      +/-   ##
==========================================
+ Coverage   82.16%   82.21%   +0.05%     
==========================================
  Files         334      334              
  Lines       19537    19537              
==========================================
+ Hits        16052    16062      +10     
+ Misses       3485     3475      -10     

Flag	Coverage Δ
ingest	69.11% <ø> (ø)
node-sidecar	95.39% <ø> (+1.38%)	⬆️
nucliadb	70.46% <ø> (+0.04%)	⬆️
reader	79.02% <ø> (ø)
sdk	40.58% <ø> (ø)
search	79.09% <ø> (ø)
standalone	88.27% <ø> (ø)
train	63.76% <ø> (ø)
utils	81.25% <ø> (ø)
writer	85.15% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

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

[javitonino]

⚠️ Behaviour for load_all differed between the sync & async version:

• The async version loaded all shards and then grabbed the lock to insert all shards into cache.
• The sync version held the lock while loading all shards.

This version behaves like the sync version and thus blocks the cache until everything is loaded. It seems safer but potentially a long block. The async version could be prone to race-conditions if a shard is opened while load_all is running. A third alternative would be to grab a write lock for each iteration of the loop, so other consumers get the chance to interact with the cache (albeit with high contention).