feat(runtime-node+core): auto-resume from checkpoint after gateway restart

[subinium]

Severity: priority/warning
Source: NousResearch/hermes-agent v0.11.0, PR #9934
Area: packages/runtime-node, packages/core, packages/storage

Problem

Hermes resumes interrupted agent work after a gateway restart from the last checkpoint instead of starting a fresh turn. CrowClaw has SessionCheckpoint storage (packages/storage/src/index.ts:517) but no auto-resume on boot.

Proposed fix

1. On runtime startup, read SessionCheckpoint for any session marked 'in_progress'.
2. Re-hydrate the session and continue from the last assistant turn / tool call.
3. Operator override: --no-resume flag.
4. Surface a session:resumed event for observability.

References

• Hermes PR #9934