[backlog] /v1/models response shape parity (upstream #507)

[DeliciousBuding]

    ## Source
    - Upstream: cita-777/metapi #507
    - Matrix: `docs/analysis/original-gap-matrix.md`
    - Category: feature-protocol
    - Status: partial · Priority: P1 · Backlog: yes
    - Epic: Protocol
    - Parent: G3 #10

    ## Evidence (from matrix)
    `handler/proxy/models.go` + e2e `e2e_flow_test.go` GET `/v1/models`; shape parity with all clients needs runtime verification.

    ## Acceptance criteria (draft — do not implement in G3)
    - [ ] `/v1/models` response shape matches OpenAI-compatible client expectations used by product

• Owned vs upstream model listing behavior documented

• e2e asserts required fields (id, object, owned_by as applicable)

• Known client-specific quirks documented

    ## Out of scope for this issue filing
    Docs-only backlog. No product code until a later implementation wave claims this issue.