Improve raw blobs downloading
What does this MR do?
This commit improves blobs downloading in two ways:
- When we send a blob via Workhorse, we don't need to download the whole file first, so we can set the limit to zero and download just enough data to detect whether it's an LFS
- When we serve a file via API, then we shouldn't load all unlimited data, we'd better have a limit there
Related issue: https://gitlab.com/gitlab-org/create-stage/-/issues/12892
Edited by Igor Drozdov