fix(stats): the collector's rate, its windows, and what happens when a VNI changes hands

[daniel-noland]

A steady load cannot test a rate pipeline, so the collector is driven with a ramp and its published rate asserted against the load that produced it. That found the rate window being read out of arrival order, startup batches that were not consecutive, a rate that could be negative or NaN, and a metric series that outlived its own name.

When a VNI changes hands its counters must reset and the traffic still in flight must be dropped; neither happened. Ends with a set of flow-entry and net fixes turned up by the same fuzzing.

---

24 commits. Stacked on pr/daniel-noland/driven-clock; review that first.

🤖 Generated with Claude Code

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

_{Comment @coderabbitai help to get the list of available commands.}

[codecov]

❌ 2 Tests Failed:

Tests completed	Failed	Passed	Skipped
1793	2	1791	0

View the top 2 failed test(s) by shortest run time

dataplane-nat::masquerade::fuzz::nothing_is_masqueraded_without_permission

Stack Traces | 1.14s run time

thread 'masquerade::fuzz::nothing_is_masqueraded_without_permission' (166568) panicked at ..../src/masquerade/fuzz.rs:192:9:
5 flows reached the permission assertion across 1 configurations; this property has gone vacuous
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

dataplane::bin/dataplane::packet_processor::fuzz::generated::a_generated_configuration_carries_its_own_traffic

Stack Traces | 1.14s run time

thread 'packet_processor::fuzz::generated::a_generated_configuration_carries_its_own_traffic' (163249) panicked at ..../src/packet_processor/fuzz.rs:865:5:
no traffic was ever derived across a peering permitting only one direction, so no reply here was authorised by the flow it belongs to and the reverse lookup in `AclFilter::lookup` was never entered. Check for a `__fuzz__` corpus beside this test before reading further: replaying one can spend the budget on inputs chosen for being unusual. Move it aside and re-run to tell that apart from a real gap.
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}