refactor(streaming): migrate OpenAI Responses SSE framing to the shared codec (phase 1 of #842)

[leseb]

Part of #842 (parent epic praxis-proxy/praxis#985).

Scope

Migrate the OpenAI Responses provider-neutral SSE record framing onto the shared praxis_filter::sse codec, and bump the Praxis dependency to a revision that carries the codec.

Consumers in scope:

• apis/src/openai/sse/frame.rs — replace the local framer with the shared SseDecoder.
• apis/src/openai/sse/responses/parser.rs — feed &Bytes chunks through the adapter.
• apis/src/openai/responses/stream_events/ — the openai_stream_events filter.

Out of scope (tracked in sibling phases)

• Anthropic stream_events framing.
• A2A route-capture scanner + token-usage reuse.

Acceptance criteria (subset of #842)

• Update the Praxis dependency to a revision containing the shared codec.
• SseFrameParser delegates byte-level line/record framing to SseDecoder; the OpenAI-facing SseFrame / SseParseError surface, [DONE] sentinel, event-count budget, timeouts, and lifecycle rules stay in the consumers.
• Preserve pass-through / transform / fail-open / fail-closed / terminal / EOF behavior; no full-body buffering; no weakened backpressure.
• Retain consumer-specific limits above the shared framing bounds.
• Remove obsolete duplicate parser tests; keep focused adapter tests plus the codec's conformance coverage.

Status

Addressed by PR #942 (Refs #842).