Add non-invasive CAVE integration spike scaffold

[akgohain]

Motivation

• Introduce a minimal, documented spike for exporting app workflow artifacts to a CAVE-compatible data flow without changing any existing runtime behavior or routes.
• Provide a clear contract and guardrails so future work can implement mappings, auth, and network logic behind an explicit opt-in path.

Description

• Add a spike document at docs/integrations/cave-integration-spike.md that explains target use cases, expected inputs/outputs, artifact-to-CAVE mapping, non-invasive guardrails, and actionable next steps.
• Add an inert adapter scaffold at server_api/workflow/cave_adapter.py with WorkflowArtifact and CavePayload dataclasses and a CaveAdapter interface whose methods explicitly raise NotImplementedError to prevent runtime side effects.
• Add contract tests at tests/test_cave_adapter_contract.py that validate dataclass shapes and assert the adapter stubs remain unimplemented.
• Add TODOs in the scaffold and doc for auth, network, and deployment assumptions to guide future implementation.

Testing

• Ran uv run pytest -q tests/test_cave_adapter_contract.py, which failed due to an environment/package metadata issue unrelated to these changes (editable dependency metadata for pytorch-connectomics).
• Ran python -m pytest -q tests/test_cave_adapter_contract.py, which succeeded with 3 passed.
• No changes were wired into startup or routes and no runtime behavior was modified by these additions.

---

Codex Task

[akgohain]

Consolidated into draft PR #165 (feat/codex-wave-integration). Closing this source draft to keep review in one place.