test: stop test_topics_filter_pulse_still_ticks flaking on interleaved pulse ticks - #148

What flaked and why

test_topics_filter_pulse_still_ticks failed on the windows and macos CI runners while passing on ubuntu and local Linux.

The actor under test is configured with pulse: 50, so it emits a filter.out frame with tick: true every 50ms. The test consumed the first tick, then appended noise and ev and asserted the recv order was exactly noise then ev:

store.append(Frame::builder("noise").build()).unwrap();
store.append(Frame::builder("ev").build()).unwrap();
assert_eq!(recver.recv().await.unwrap().topic, "noise"); // got "filter.out" on slow runners
assert_eq!(recver.recv().await.unwrap().topic, "ev");

On a slow runner, more than 50ms passed between consuming the first tick and this recv, so a second pulse tick landed in the store ahead of noise and the strict assert_eq!(... "noise") got filter.out instead. The actor is behaving correctly: async pulse ticks interleave with data frames, and the order between a tick and a data frame is not guaranteed. The loop further down already tolerated interleaved ticks; these two asserts did not.

Deterministic repro

Sleep 60ms (past one 50ms pulse period) after consuming the first tick and before appending, so a second tick is always queued ahead of the data frames on any machine. With that in place the old strict asserts fail deterministically (assertion left == right failed, left: "filter.out", right: "noise"), matching CI.

The fix

Keep the forced interleave so the test genuinely exercises the tick-between-data-frames case going forward, and tolerate ticks the same way the loop already did: drain filter.out tick frames, then assert only that noise and ev arrive in append order and the actor's filter.out {seen: ev} output arrives.

Verification

• cargo test: 189 + 12 + 8 passed, 0 failed
• processor::actor::tests module: 35 passed
• cargo fmt --check and cargo clippy -- -D warnings: clean
• nu tests/test_xs_nu.nu: a pre-existing scru128 id-roundtrip failure reproduces on a clean checkout of this branch's base too, unrelated to this change
• docs/astro step in check.sh needs Node >=22.12.0; the local Node is v18.19.1, so that step is gated by the environment (unrelated to this change)