Stream GET replies

[roman-khimov]

Is your feature request related to a problem? Please describe.

I'm always frustrated when we're wasting time and memory for nothing. Current GET (and GETRANGE) implementation does complete buffering for both local and networked requests.

Describe the solution you'd like

Network API always had a concept of payload stream, FSTree can also provide it now after #3431. The only thing needed is to make it available from shard/local engine API and use it in object service. Play with buffers a bit (64K should be OK after #3406 and SDK version update).

Describe alternatives you've considered

Keep pulling megabytes of data into RAM, wait for this data to be read completely before replying to user.

[roman-khimov]

If you're to try doing

$ neofs-cli get --cid 411d8vuzogogMxXJqTQcu61btgQ6rL2VNYUYnH7r4kE3 --oid 4QNXbGzU3ooJgpsR7EVawyKgtrQSDDw5BwdLYeND9gZT -r grpc://st3.storage.fs.neo.org:8080 --timeout 1000s --file Phase1_2

you can clearly see the process pausing on 64M boundaries, bytes flow in, then there is an interruption, then they start coming in again. That's the effect of buffering as well (plus a new request, but that's a different problem).