Expose elapsed_ms in step.http_call output for latency measurement

[intel352]

Summary

step.http_call does not currently expose the request duration in its output. We need an elapsed_ms field in the step output so pipelines can measure and log upstream latency.

Use Case

In a facade/proxy pattern with shadow mode, a pipeline calls two upstream services in parallel and logs divergence metrics including latency:

- name: call_primary
  type: step.http_call
  config:
    url: "https://primary-service/api/resource"
    method: GET

- name: call_shadow
  type: step.http_call
  config:
    url: "https://shadow-service/api/resource"
    method: GET

- name: log_divergence
  type: step.log
  config:
    level: info
    message: "primary_latency={{.steps.call_primary.elapsed_ms}}ms shadow_latency={{.steps.call_shadow.elapsed_ms}}ms"

Currently there is no way to access the request duration from step.http_call output, forcing workarounds like -1 sentinel values in latency metrics.

Proposed Behavior

• Add elapsed_ms (int64, milliseconds) to step.http_call output alongside existing fields (status_code, body, headers)
• Measured as wall-clock time from request send to response fully read
• Available in pipeline context as steps.<step_name>.elapsed_ms

Context

This is a blocker for enabling shadow mode divergence logging in a multi-service migration. Without real latency values, divergence metrics are meaningless for capacity planning and performance comparison between services.