feat(serve): describe endpoint traffic as distributions over any window

Address review feedback on describe_endpoint_traffic (renamed from
read_endpoint_traffic) and Workload.from_endpoint:

- The window is relative (lookback back from end_time, seven days by
  default) or absolute (start_time to end_time). Times must be
  timezone-aware and are reported in UTC.
- Input tokens, output tokens and concurrency are TrafficStats: min, p25,
  p50, p70, p80, p90, p95, p99, max (p100), mean and standard deviation.
  Token percentiles interpolate within the histograms' buckets.
  Concurrency's are over the window's hourly maximums, so its max is the
  busiest hour's.
- Token histograms are listed per serving framework, so supporting
  another framework is one entry, and the result names the framework
  read. vLLM's are listed today.
- Token metrics aren't read when the endpoint configuration turns
  detailed observability off, and the result reports the setting.
  Concurrency doesn't depend on it.
- The result includes the variant's Data Capture prefix.
  Workload.from_endpoint replays the captured requests when the endpoint
  captures them (use_data_capture=False opts out), and otherwise builds a
  synthetic workload from the token metrics.
- Workload.from_endpoint passes tokenizer, hf_token and the other
  workload parameters through **params.

Author: ZealSV

sagemaker-serve/src/sagemaker/serve/ai_inference_recommender/__init__.py

@@ -41,8 +41,8 @@
 from sagemaker.serve.ai_inference_recommender.secrets import Secret
 from sagemaker.serve.ai_inference_recommender.traffic import (
     EndpointTraffic,
-    TokenStats,
-    read_endpoint_traffic,
+    TrafficStats,
+    describe_endpoint_traffic,
 )
 from sagemaker.serve.ai_inference_recommender.workload import Workload
 from sagemaker.serve.ai_inference_recommender._model_builder_methods import (
@@ -62,12 +62,12 @@
     "PerformanceTarget",
     "RecommendationJob",
     "Secret",
-    "TokenStats",
+    "TrafficStats",
     "Workload",
     "WorkloadValidationError",
     "compare_benchmarks",
+    "describe_endpoint_traffic",
     "list_benchmarks",
     "list_recommendations",
-    "read_endpoint_traffic",
     "start_benchmark",
 ]