fix: retry transient backend errors across storage handlers and metric sync

[aditya-neuraco]

No description provided.

[damonhayhurst-neuraco]

I find this confusing.

1. The consts you've added in core imply that we retry on these status calls for all backend api calls.
2. The redirection away from thread_local_session() leads to a misleading read.
3. You've localised some retry mechanisms to a locally named mixin that should really be generic.

My suggestion would be to extend thread_local_session() with retry mechanisms that can be overriden when necessary.