* Set transaction_per_migration=True

* Document Server upgrades

Signed-off-by: Matías Insaurralde <[email protected]>

…context (#3226)

Signed-off-by: Matías Insaurralde <[email protected]>

Fixes: #3225

* Drop gerrors in runner.go

* Drop gerrors in http.go

* Drop gerrors in repo/manager.go

* Drop gerrors in executor/files.go

* Drop more gerrors usages

* Drop gerrors in executor/repo.go

* Drop gerrors in executor/executor.go

* Drop gerrors package

* Drop tracerr

* Drop tracerr from go.mod

* Fix wrapping nil errors

* Wrap errors

* Propagate check dstack-runner exists errors

* Wrap errors in executor.go

* Fix runpod type annotations

* Add _generate_create_cluster_mutation

* feat: add create_cluster method to RunpodApiClient

* feat: add delete_cluster method to RunpodApiClient

* Use keyword arguments

* Implement run_jobs and terminate_compute_group

* Prototype compute.run_jobs calling

* Prototype compute group provisioning for multinode tasks

* Add JobModel.waiting_master_job

* Add ComputeGroupModel

* Implement process_compute_groups to terminate compute groups

* Remove todo

* Fix comments

* Set internal_ip

* Support Runpod Clusters offers

* Respect supported pod_counts

* Support registry_auth

* Fix tests

* Add feature flag DSTACK_FF_RUNPOD_CLUSTER_OFFERS_ENABLED

* Remove hardcoded template_id

* Use BACKENDS_WITH_GROUP_PROVISIONING_SUPPORT

* Switch to nebius sdk 0.3

* Pass auth_options to op.update

* Do not terminate fleet instances on idle_duration at nodes.min

* Update idle_duration reference

* Skip nodes.min check for autocreated fleets

Reduces log noise when DSTACK_SHIM_LOG_LEVEL is set to 5 or lower

Use the fabrics from gpuhunt offers instead. The
list of fabrics was moved from dstack to gpuhunt
so that new fabrics can be added without a dstack
release.

Fixes: #3236