The deleted wrapper put cache: {"no-cache": true} on every request body, so the
gateway's response cache never answered a re-sent prompt. The SDKs send nothing
of the sort, and the mid-conversation prompt-cache priming loop re-sends an
identical body until the provider reports a warm cache, which a cached reply
never does. NO_PROXY_CACHE in sdk_clients.py restores the field as extra_body
on every messages, responses, completions, and embeddings call.
The wrapper also returned a 4xx as a value where the SDKs raise. The Bedrock
safety_identifier test judges the captured Converse body, and Claude on Bedrock
rejects the forwarded field with a 400, so the /v1/responses leg now suppresses
openai.BadRequestError the way the chat leg carries the same 400 as a Result.
The llm_translation suite drove /v1/responses, /v1/messages, /embeddings,
/v1/images/generations, /v1/moderations and /v1/audio/* through a bespoke
endpoints_client wrapper that no customer runs. Tests now call the proxy the
way customers do: the OpenAI SDK for the OpenAI-compatible surface and the
Anthropic SDK for /v1/messages, wired through a session-scoped sdk fixture
(sdk_clients.py) that points both SDKs at the proxy with a virtual key.
Endpoints no official SDK covers keep the shared typed transport: rerank moves
onto ProxyClient (RerankBody/RerankResponse in models.py) and the passthrough
header test parses with the shared AnthropicMessagesResponse model. Files that
only used endpoints_client for model registration now use the proxy fixture
directly. endpoints_client.py is deleted; anthropic joins the e2e-dev
dependency group so the lint env resolves the SDK imports.
Resolves LIT-4577
