test(data): make the push/read race test deterministic

The completion check was scheduling-dependent and the test could lose its TSan

overlap window. The producer now signals done with a release store, the consumer

reads in a do/while keyed on that flag so at least one read is guaranteed by

construction, and a started/reading handshake makes the producer's remaining

pushes provably overlap the reader's loop. Bounded to 4096 pushes on the 256-slot

ring, and the assertions moved after join() so a failure reports instead of

terminating. With the pre-fix ring restored, four TSan runs still flag push vs

copy_from, so the case keeps its teeth.

tests/test_series.cpp

@@ -2,6 +2,7 @@
 #include <plot/data/array_series.hpp>
 #include <plot/data/stream_series.hpp>
 #include <algorithm>
+#include <atomic>
 #include <cmath>
 #include <random>
 #include <thread>
@@ -104,27 +105,57 @@ PLOT_TEST(stream_series_exact_queries_survive_ring_wrap) {
 PLOT_TEST(stream_series_push_and_read_are_race_free) {
   // The ring overwrites, so the producer reuses the slot of the snapshot's oldest
   // retained sample while the consumer may still be reading it. Relaxed atomic
-  // slots make that defined; nothing else in the suite has two threads, so this
-  // is the case ThreadSanitizer actually flags when that is not the case.
-  plot::data::StreamSeries s(1u << 8, 0.0, 1.0);
-  constexpr u64 kPushes = 50'000;
-  std::thread producer([&s] {
-    for (u64 i = 0; i < kPushes; ++i) s.push(f32(i % 971));
+  // slots make that defined; nothing else in the suite has two threads, so this is
+  // the case ThreadSanitizer actually flags when the slots are plain.
+  //
+  // The two flags make the overlap structural, not scheduling luck: the reader
+  // waits until the producer is already pushing, acknowledges, and only then does
+  // the producer push the rest -- so the reader's window spans pushes that are in
+  // flight by construction. The reader then loops until the producer reports done,
+  // which also makes "at least one read" true by construction instead of by
+  // winning a race with a fast producer.
+  plot::data::StreamSeries s(1u << 8, 0.0, 1.0);   // one read spans reused slots
+  constexpr u64 kStride = 512;                     // pushes before the reader starts
+  constexpr u64 kPushes = 4'096;                   // total: a few thousand only
+  std::atomic<bool> started{false};
+  std::atomic<bool> reading{false};
+  std::atomic<bool> done{false};
+
+  std::thread producer([&] {
+    for (u64 i = 0; i < kStride; ++i) s.push(f32(i % 971));
+    started.store(true, std::memory_order_release);
+    while (!reading.load(std::memory_order_acquire)) {
+    }
+    for (u64 i = kStride; i < kPushes; ++i) s.push(f32(i % 971));
+    done.store(true, std::memory_order_release);
   });
+
+  while (!started.load(std::memory_order_acquire)) {
+  }
+  reading.store(true, std::memory_order_release);
+
   u64 reads = 0;
-  while (s.produced() < kPushes) {
+  bool sizes_ok = true;
+  bool finished = false;
+  do {
+    finished = done.load(std::memory_order_acquire);
     s.acquire();
     const u64 n = s.size();
-    if (n == 0) continue;
-    (void)s.reduce(0, n);
-    (void)s.sum(0, n);
-    (void)s.sumsq(0, n);
-    std::vector<f32> out;
-    s.gather(0, 0, n, out);                             // reads the oldest edge
-    PLOT_CHECK(out.size() == 2 * n);
+    if (n > 0) {
+      (void)s.reduce(0, n);
+      (void)s.sum(0, n);
+      (void)s.sumsq(0, n);
+      std::vector<f32> out;
+      s.gather(0, 0, n, out);                           // reads the oldest edge
+      if (out.size() != 2 * n) sizes_ok = false;
+    }
     ++reads;
-  }
+  } while (!finished);
+
   producer.join();
+  // Checked only after the join: a throw while the producer is joinable would
+  // std::terminate instead of reporting the failure.
   PLOT_CHECK(s.produced() == kPushes);
-  PLOT_CHECK(reads > 0);
+  PLOT_CHECK(reads >= 1);
+  PLOT_CHECK(sizes_ok);
 }