Better sampling override behavior

[trask]

(icm 577578458)

Currently, this configuration:

      {
        "telemetryType": "request",
        "attributes": [
          {
            "key": "http.url",
            "value": ".*/health-check",
            "matchType": "regexp"
          }
        ],
        "percentage": 0,
        "id": "filter out health check"
      },
      {
        "telemetryType": "dependency",
        "attributes": [
          {
            "key": "db.statement",
            "value": "select count(*) from abc",
            "matchType": "strict"
          }
        ],
        "percentage": 50,
        "id": "filter out noisy jdbc"
      }

has the unexpected effect of capturing 50% of select count(*) from abc calls, even when they occur within the excluded health check.

this PR updates the behavior so that when a sampling override percentage is <100 and a parent dropped, then the span (with override percentage < 100) will also be dropped

TODO need to decide whether this affects statistical accuracy of item count

THOUGHT when parent span isn't sampled, it's a no-op, and so has no "sample rate" attribute available for children to check