feat: timing spans

[noahpodgurski]

What's new

• Add support for returning span timing data in any MLPA response. Simply add header debug-timing-key: <env.MLPA_DEBUG_TIMING_KEY> to the request.
• RequestTimingsMiddleware is a new middleware which adds request profiling. It wraps the ASGI send callable and inspects http.response.start for response headers, then handles http.response.body messages: JSON bodies are collected, spans are added, and sent with an updated Content-Length. Streaming SSE chunks pass through immediately, with spans appended when the final DONE body message arrives.
• Timing format returns at the top level of the response (for non streaming) and for successful streams, after data: [DONE], receive one additional data: {"spans": [...]} event.

Examples:

Non streaming

"spans": [
    {
      "name": "total",
      "start_ms": 0.0,
      "duration_ms": 717.08
    },
    {
      "name": "auth",
      "start_ms": 0.84,
      "duration_ms": 0.52
    },
    {
      "name": "traffic_contract",
      "start_ms": 1.38,
      "duration_ms": 0.0
    },
    {
      "name": "db",
      "start_ms": 1.38,
      "duration_ms": 3.06
    },
    {
      "name": "upstream",
      "start_ms": 4.5,
      "duration_ms": 711.26
    }
]

Streaming

data: {"spans": [{"name": "total", "start_ms": 0.0, "duration_ms": 1311.5}, {"name": "auth", "start_ms": 1.06, "duration_ms": 238.21}, {"name": "traffic_contract", "start_ms": 239.65, "duration_ms": 0.01}, {"name": "db", "start_ms": 239.79, "duration_ms": 19.7}, {"name": "upstream", "start_ms": 259.87, "duration_ms": 1051.05}]}

https://mozilla-hub.atlassian.net/browse/AIPLAT-1287

[noahpodgurski]

/build-pr