Design doc for the fluent query core semantics, before implementation:

• A chain is a lazy query plan; evaluation happens only at terminators.
• Operates on canonical models (Epic A): URIs in, URIs out.
• Backend planning: which stages execute in-process vs compile to Cypher.
• Caching semantics for repeated chains (the analysis cache already exists; query results should follow).
• Error/empty-result semantics; how .explain() records the plan, backend resolutions, and the unresolved-edge set.

Updates §1/§4 of the epic proposal (#155) for the canonical model.

Part of #155. Branch: chore/issue-<n>.