Track project import lifecycle internal events

What

Fires internal events for the project import lifecycle, so every import transition can be tracked in real time regardless of importer, using the event names and properties agreed with the Data team in #617884.

Five events are emitted across the lifecycle:

• start_project_import
• finish_project_import
• fail_project_import
• cancel_project_import
• timeout_project_import

Two import paths are covered:

Standard importers — app/models/project_import_state.rb, via after_transition hooks. Covers GitHub, Bitbucket, Bitbucket Server, Gitea, Fogbugz, file-based (gitlab_project), Manifest, and Git/Repository-by-URL.

• project.mirror? is skipped: ProjectImportState is backed by the project_mirror_data table and its state machine is reused by recurring pull-mirror updates, so without this guard every mirror sync would be counted as an import.
• Only Gitlab::ImportSources.importable_project_types are tracked, which excludes template-based creation (gitlab_built_in_project_template).
• mark_as_timed_out sets a transient timed_out flag and transitions to failed via the existing fail_op event, same as mark_as_failed — no new state. ProjectImportState is shared with recurring pull-mirror updates, and the mirror recovery logic (CE's schedule event, EE's retry/capacity bookkeeping) only knows about failed, so a genuinely new timeout state would break mirror recovery. The after_transition :failed handler picks the tracked event name (timeout_project_import vs fail_project_import) from the flag instead of the target state. Gitlab::Import::StuckProjectImportJobsWorker now flags an import as timed out instead of just failed; other StuckImportJob includers (Jira) are unaffected.

Direct Transfer — Bulk Imports:

• lib/bulk_imports/projects/pipelines/project_pipeline.rb — started when the project record is created.
• lib/bulk_imports/common/pipelines/entity_finisher.rb — finished when a project_entity reaches the finished state.
• app/models/bulk_imports/entity.rb — failed, cancelled, and timeout (cleanup_stale).

BulkImports::Entity and EntityFinisher are shared with Offline Transfer, which imports projects through its own, nearly identical pipeline (lib/import/offline/projects/pipelines/project_pipeline.rb). Rather than excluding Offline Transfer, that pipeline now emits start_project_import too, so both transfer mechanisms are covered by the same five events. The two are distinguished by label, which is read from project.import_type (gitlab_project_migration for Direct Transfer, offline_transfer for Offline Transfer) instead of a hardcoded string. The shared project/user/namespace/additional_properties hash is extracted into BulkImports::Entity#project_import_event_attributes, reused by ProjectPipeline (both DT and Offline Transfer), EntityFinisher, and Entity itself.

Every event carries project / user / namespace identifiers, a label of the importer name, and a hashed import_source property (Gitlab::Import::SourceIdentifier, wrapping Gitlab::CryptoHelper.sha256) so repeated imports of the same source can be detected without sending the raw host/path to Snowplow (import_source counts as Red data under the Data Classification Standard).

Scope notes

• This MR covers #627868 (closed) (the import_source hashing helper) and #627872 (closed) (the standard-importer + Direct Transfer project lifecycle events); it is the first of several planned slices of #617884. request_channel, source_hosting, imported_objects_count, and fidelity_rate are deferred to follow-up MRs (tracked as separate child tasks under #617884).
• Builds on the closed !236324 (closed), whose author intentionally deferred it to consolidate under #617884.

References

• #617884
• #627868 (closed)
• #627872 (closed)
• !236324 (closed)

How to set up and validate locally

1. Enable Snowplow Micro so events are actually collected:

   gdk config set snowplow_micro.enabled true
   gdk reconfigure
   gdk restart

2. Import a Project

   • New project
   • Import project

3. Import a Project using Direct Transfer

   • New group

4. Observe events in Snowplow Micro (http://gdk.test:9091/micro/good)