Add non-invasive CAVE integration spike (docs + adapter scaffold + tests)

[akgohain]

Motivation

• Provide a lightweight, non-invasive spike that documents how this app could integrate with CAVE-compatible pipelines without changing runtime behavior.
• Surface input/output contracts, mapping rules, and deployment/auth/network TODOs so implementation can proceed safely behind a non-default path.

Description

• Add a spike document at docs/integrations/cave-integration-spike.md that describes target use case, adapter input/output contracts, artifact→CAVE concept mapping, proposed opt-in flow, TODOs for auth/network/deploy, and actionable next steps.
• Add an inert adapter scaffold at server_api/workflow/cave_adapter.py defining typed dataclasses (WorkflowArtifact, CavePayload, CaveResult) and an interface class CaveWorkflowAdapter whose methods build_payload and parse_result deliberately raise NotImplementedError.
• Add contract tests in tests/test_cave_adapter_contract.py that explicitly load the adapter module to verify it produces no side effects on import and that adapter methods remain stubbed.
• Ensure the adapter is opt-in and does not alter any existing routes, startup behavior, or runtime paths.

Testing

• Ran python -m pytest -q tests/test_cave_adapter_contract.py which passed (2 tests).
• Running the project-scoped command uv run pytest -q tests/test_cave_adapter_contract.py failed in this environment due to a packaging/metadata resolution issue for an editable dependency (pytorch-connectomics), which is unrelated to the spike code.
• The added tests assert explicit module import without side effects and that build_payload/parse_result raise NotImplementedError as expected.

---

Codex Task

[akgohain]

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