Add SSE/chunked streaming support to the SDK

What

Add SSE / chunked streaming support to call_runner in the Python SDK, so callers can consume live runner responses incrementally instead of waiting for the full body.

• http.open_stream() — opens an HTTP request and hands back the live (session, response) without reading the body. Caller owns both. No total timeout (streams run indefinitely); only connect/first-byte are bounded. Raises LivepeerHTTPError on >= 400.
• LiveRunnerCallStream — returned by call_runner(..., stream=True). Owns the aiohttp session/response, is an async context manager (or call aclose()), and exposes the body via aiter_bytes() / aiter_lines() plus status/headers/content_type/runner.
• Payment — streaming reuses the existing per-call 402 challenge/retry path; open_stream raises before any body, so the payment retry catches it exactly like the non-streaming path.
• Export LiveRunnerCallStream from the package root.

Why

Live runner apps that stream output (token-by-token LLM responses, progressive media) need an incremental response path in the SDK, without standing up the trickle/lv2v continuous-payment flow. This SSE path is one job, one upfront payment.

Implementation

Implemented in livepeer/livepeer-python-gateway#25 (builds on the live-runner work in #20).