metrics tests

- Add _create_metric_stub_from_protobuf method to correctly parse protobuf metrics
- Add _extract_attributes_from_data_point helper method
- Change metric handling to use protobuf-specific parsing instead of OpenTelemetry native parsing
- Add missing typing import
- Add OTEL_METRIC_EXPORT_INTERVAL environment variable for test configuration

This fixes the CI failure where metrics were not being properly extracted from
protobuf data in server mode tests.

tests/integration/telemetry/recordings/1fcfd86d8111374dc852cfdea6bfdb6a511f92cee84a6325b04ae84878512c30.json

@@ -30,7 +30,7 @@
             "index": 0,
             "logprobs": null,
             "message": {
-              "content": "import torch\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\n\n# Load the pre-trained model and tokenizer\nmodel_name = \"CompVis/transformers-base-uncased\"\nmodel = AutoModelForCausalLM.from_pretrained(model_name)\ntokenizer = AutoTokenizer.from_pretrained(model_name)\n\n# Set the temperature to 0.7\ntemperature = 0.7\n\n# Define a function to generate text\ndef generate_text(prompt, max_length=100):\n    input",
+              "content": "To test the trace function from OpenAI's API with a temperature of 0.7, you can use the following Python code:\n\n```python\nimport json\n\n# Import the required libraries\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\n\n# Set the API endpoint and model name\nmodel_name = \"dalle-mini\"\n\n# Initialize the model and tokenizer\nmodel = AutoModelForCausalLM.from_pretrained(model_name)\ntokenizer = AutoTokenizer.from_pretrained(model_name)\n\n",
               "refusal": null,
               "role": "assistant",
               "annotations": null,
@@ -55,5 +55,6 @@
       }
     },
     "is_streaming": false
-  }
+  },
+  "id_normalization_mapping": {}
 }