Centralize full environment export for stopped runs

[emmepra]

Problem / Opportunity

For run environments with meaningful internal state, full environment-side export is still too dependent on the run stack staying alive. Summary-level recovery is not enough when a user wants the full environment artifact set from a stopped run, such as complete inspector events, full post snapshots, or other environment-owned state.

The platform should preserve these artifacts centrally at terminalization instead of requiring ad hoc stack relaunches or direct access to environment-local volumes.

Desired Outcome

A stopped run should support durable full export of environment-owned run data through controller-managed storage, without requiring the original run containers to still be running.

Proposed Direction

• Define a controller-owned terminal export contract for environment stacks.
• Before stop/delete, capture and persist full environment export artifacts (for example inspector events, full post snapshot, and any environment-defined state bundle).
• Store these artifacts centrally and expose them through stable controller/admin export surfaces.
• Keep relaunch-based recovery only as a temporary backfill path, not the normal export path.

Scope

• In scope: durable full export for stopped runs and admin/reporting workflows.
• Out of scope: redesigning every environment’s live inspector UX.

Acceptance Criteria

• A stopped run can export full environment-owned artifacts without relaunching the run stack.
• A deleted run remains exportable if its terminal artifacts were already centralized.
• The export contract is generic enough for multiple environments, not only a single private study.
• Study/export scripts can obtain full run artifacts from controller/admin APIs without calling the per-run frontend directly.

Context

• Source: private study analysis needs showed that environment-local state survives in run volumes, but post-hoc export should not rely on live stack availability.
• Related: controller events export, per-run environment state, orchestrator stop/delete lifecycle.