What does this PR do?

When creating a new metric context, the rate parameter is currently not passed to the metric constructor. I believe that means sampling rates for histogram/distribution/timing metrics are ignored. It also means that the parameters we did pass to the constructor were shifted by one position, making them do not what we wanted.

Description of the Change

This adds forwarding of the rate parameter to the constructor we call, and adds a unit test to ensure forwarding is done properly.

Possible Drawbacks

I don't see any.

Reproducer

This is as simple as I could make it -- if we pass parameters properly, the "max samples per context" limiting should work.

from unittest.mock import patch
from datadog.dogstatsd.max_sample_metric import HistogramMetric
from datadog.dogstatsd.max_sample_metric_context import MaxSampleMetricContexts

MAX_SAMPLES = 10

with patch("datadog.dogstatsd.max_sample_metric_context.random.random", return_value=0.0):
    ctx = MaxSampleMetricContexts(HistogramMetric)
    for _ in range(1000):
        ctx.sample("m", 1.0, ["t:v"], 0.5, "key", max_samples_per_context=MAX_SAMPLES)

metric = ctx.values["key"]
print(f"buffer size: {len(metric.data)} (expected: {MAX_SAMPLES})")
assert len(metric.data) == MAX_SAMPLES, f"BUG: buffer grew to {len(metric.data)}"

Additional Notes

❓ I think this problem is one that would typically be detected and prevented by even basic static typing / static correctness checking. Has anyone ever looked into this, and would you be opposed to (me) adding some? Not necessarily a full blown typing of everything, but at least some basics.