MCPJVM-366: SQLite state-store rebuild from canonical Artifacts

[nimbly-dev]

Description

The primary long-term recovery path for the SQLite-backed state-store should be deterministic rebuild from canonical per-run Artifacts. Since execution.result.json, evidence.json, correlation.json, and related run Artifacts represent the source of truth, the state-store should be able to re-derive its persisted rows from those Artifacts without depending on legacy shared indexes.

This story adds canonical Artifact-based rebuild behavior for the SQLite-backed state-store.

Scope

In scope:

• Rebuild SQLite-backed state-store content from canonical per-run Artifacts
• Support deterministic re-derivation of persisted run and correlation state from existing Artifact evidence
• Preserve canonical Artifacts as the source of truth during rebuild operations
• Keep rebuild behavior fail-closed when required Artifact evidence is missing or invalid
• Keep implementation aligned with the existing artifact_management action-plus-shared helper structure

Out of scope:

• Long-term support for legacy shared correlation-index.json import
• New public MCP query/read contracts beyond the rebuild capability entrypoint
• Replacing canonical Artifacts with the DB itself
• Performance Suite integration

Expected Tables / Detailed Schema Direction

Rebuild must be able to repopulate at least these targets from canonical evidence:

Table	Rebuild source	Minimum direction
suite_runs	execution.result.json, context.resolved.json, run directory metadata	restore run identity, status, timing, Artifact linkage
correlation_runs	correlation/correlation.json	restore key/session/window/verdict summary
watcher_runs	watcher-related canonical evidence when present	restore watcher status/outcome summary
external_verifications	canonical verification evidence when present	restore verification status/provider summary
artifacts	canonical Artifact file set	restore Artifact link rows

Acceptance Criteria

• SQLite-backed state can be rebuilt deterministically from canonical per-run Artifacts
• Rebuild uses Artifact evidence as the source of truth rather than legacy shared indexes
• Rebuild fails closed when required deterministic evidence is missing or invalid
• The implementation remains aligned with the existing artifact_management action-plus-shared helper structure
• The rebuilt state is sufficient to restore persisted run/correlation indexing behavior

Implementation Notes

• Rebuild should be the preferred long-term maintenance path and should not depend on legacy JSON index presence.
• Keep rebuild logic bounded and summary-oriented in its public response even if internal scanning covers many runs.

Testing

• UT for Artifact-to-state rebuild mapping logic
• UT for fail-closed handling on missing or invalid canonical Artifacts
• IT for full rebuild from existing regression run Artifacts
• IT for deterministic rebuild summaries on repeated runs over the same Artifact set