`datadog_metrics`: sink fails with `Some(BadRequest)` and drops events

Problem

The datadog_metrics sink fails with Some(BadRequest) and drops the events in the request:

ERROR Service call failed. No retries or retries exhausted.
      error=Some(BadRequest) error_type="request_failed" stage="sending"
      component_kind="sink" component_type="datadog_metrics"
ERROR Events dropped intentional=false count=1
      reason="Service call failed. No retries or retries exhausted."

Version

0.56.0

Reproduction

Expose a histogram whose observations all fall above its top bucket, so every one of them ends up
in the implicit +Inf bucket (pip install prometheus_client):

# fixture.py
import time

from prometheus_client import CollectorRegistry, Histogram, start_http_server

registry = CollectorRegistry()
# The only bucket is le=1, and every observation is far above it.
h = Histogram("test_overflow", "demo", buckets=(1.0,), registry=registry)

start_http_server(19092, registry=registry)
while True:
    h.observe(5000)
    time.sleep(1)

which exposes:

test_overflow_bucket{le="1.0"} 0.0
test_overflow_bucket{le="+Inf"} 4.0
test_overflow_count 4.0
test_overflow_sum 20000.0

# repro.yaml
sources:
  prom:
    type: prometheus_scrape
    endpoints: ["http://127.0.0.1:19092/metrics"]
    scrape_interval_secs: 3

sinks:
  dd:
    type: datadog_metrics
    inputs: [prom]
    default_api_key: "${DD_API_KEY}"

python3 fixture.py &
DD_API_KEY=<key> vector --config repro.yaml

Every scrape after the first produces one such failure.

Start by running fixture.py with repro.yaml and observe the datadog_metrics sink handling of the Prometheus +Inf bucket. Trace the sink's request and error path for histogram observations above the top bucket; done means repeated scrapes no longer produce Some(BadRequest) or drop events.