Description

When an AI provider's SSE stream stalls mid-response (e.g. the upstream sends a
few chunks then hangs), the body_reader() call in parse_streaming_response
times out. The error handler at ai-providers/base.lua:551 unconditionally
returns a transport-layer error code via transport_http.handle_error(err)
(504 for timeout, 500 for other errors).

However, before entering parse_streaming_response, the caller at
ai-proxy/base.lua:350 has already set the Content-Type response header.
If at least one SSE chunk has been dispatched to the downstream client, the
HTTP response headers — including status 200 — have already been flushed.
Returning a 504/500 after this point causes nginx to emit:
attempt to set status 504 via ngx.exit after sending out the response status 200
attempt to set ngx.status after sending out response headers

Checklist

• I have explained the need for this PR and the problem it solves
• I have explained the changes or the new features added to this PR
• I have added tests corresponding to this change
• I have verified that this change is backward compatible (If not, please discuss on the APISIX mailing list first)