Serve graph endpoints through the inspection adapter

[jitu5]

Parent issue: #2265
Depends on: #2657, #2658, #2687, #2715

Update: The original ticket described an environment flag, two graph providers, live fallback, and a special live path for --params. Those decisions are obsolete. The inspection adapter will be the only engine serving runtime HTTP graph responses, and resolved runtime parameters come from the existing config loader.

Summary

Serve /api/main and /api/pipelines/{id} from the snapshot GraphBuilder. The adapter is already implemented and tested, but these routes still use the old data_access_manager graph path.

Scope

• Add one small inspection-adapter provider around GraphBuilder.
• Build and install the provider once during server startup using the selected project, environment, pipeline, catalog configuration, and resolved parameters.
• Preserve the full graph response contract before cutover: attach .viz stats/styles as node_extras, and in full mode use a same-ID live lookup to supply the resolved dataset_type where the snapshot contains only a raw catalog type string. This enrichment must not change graph topology or IDs.
• Route /api/main and /api/pipelines/{id} through that provider.
• Honour --env, --pipeline, and --params.
• Fail clearly if the adapter cannot be built; do not silently switch to a second graph engine.

At this point the adapter is authoritative only for runtime HTTP graph endpoints. The current live load may remain temporarily for full-mode graph enrichment, node metadata/previews, static export, and notebook/VSCode consumers. Moving /api/nodes/{id}, /api/run-status, and static export is #2660; skipping the live load is #2661; deleting old graph construction is a later cleanup task.

Out of scope

• An adapter feature flag.
• LiveDataProvider or automatic live-graph fallback.
• Node metadata, static export, notebook/VSCode migration, and lite startup.

Acceptance criteria

• /api/main and every valid /api/pipelines/{id} response come from the inspection adapter.
• Responses match the captured baseline, including exact IDs, task parameters, node_extras, and resolved dataset types.
• Existing graph styles, stats, and dataset-type icons do not regress.
• The requested pipeline and environment are honoured.
• Typed runtime parameter overrides appear in task-node parameters without changing graph IDs.
• An unknown pipeline ID returns 404.
• Provider construction occurs once per server load, not once per request.

[ravi-kumar-pilla]

Hi @jitu5 ,

I did not understand this ticket. Lets catchup when you are available. Thank you