Generate sync clients from async ones to remove duplicated code

[vdusek]

Sync and async clients are hand-maintained copies (~6,300 lines each side). Every change is made twice and the copies already drift (method order, formatting). Mechanically un-asyncing the async classes leaves only ~500 differing lines, mostly in code that is genuinely different: HTTP clients, StreamedLog, StatusMessageWatcher, RQ batch_add_requests.

Proposal: write async by hand, generate sync from it, commit both, and fail CI when they're out of sync. This is what httpcore, psycopg, pymongo and elasticsearch-py do. Tooling: unasync, unasyncd, or a small script in scripts/. The divergent parts listed above stay hand-written.

Alternatives are worse: sharing more logic in ResourceClientBase still duplicates signatures and docstrings, and wrapping async in an event loop for sync breaks inside running loops.

✍️ Drafted by Claude Code

[vdusek]

Just an idea, not really sure about it, and it certainly has its drawbacks. We also already have the current setup in place, and our sync and async code differ in behavior in a few places (wait_for_finish, batch_add_requests with max_parallel supported only in async, StreamedLog and StatusMessageWatcher using threads vs tasks, the HTTP client implementations).