Faster piece getting via RPC

[nazar-pc]

JSON decoding is apparently quite costly for pieces, so new archived segment causes head-of-line blocking on RPC connection and apparently this results in some subscriptions to silently break somewhere in jsonrpsee (see paritytech/jsonrpsee#1409). Not sure what breaks, where or why, but this change makes a lot of sense either way.

Before:

2024-06-15T19:08:06.603299Z  INFO subspace_farmer::farmer_cache: Synchronizing piece cache
2024-06-15T19:08:08.105973Z  INFO subspace_farmer::farmer_cache: Finished piece cache synchronization

After:

2024-06-15T19:04:43.315030Z  INFO subspace_farmer::farmer_cache: Synchronizing piece cache
2024-06-15T19:04:43.793678Z  INFO subspace_farmer::farmer_cache: Finished piece cache synchronization

So even with fresh farmer in dev network initial piece cache sync is measurably faster. We'll see if this is sufficient, didn't want to use tokio::task::spawn_blocking here (though it is an option if we want to offload decoding compute out of async code completely).

Code contributor checklist:

• I have read, understood and followed contributing guide

[nazar-pc]

Restored 10 pieces at a time limit because subscriptions still break with higher decoding concurrency, even though less frequently.

The base branch was changed.