querylog: the writer queue defaults to 1024 entries, and a live stream allocates its ring on subscribe instead of at boot

src/storage/logger.zig

@@ -268,7 +268,7 @@ pub const Entry = struct {
 
 /// The memory budget the queue is allowed to occupy. `Entry` travels by value,
 /// so the composition root allocates `query_log_buffer_max` of them in full at
-/// boot (`app.zig`) and the SSE hub embeds a ring of them per subscriber.
+/// boot (`app.zig`), and each live SSE subscriber holds a ring of them.
 const queue_budget_bytes = 64 * 1024 * 1024;
 
 /// The ceiling `config/validate.zig` enforces on `logging.query_log_buffer_max`,
@@ -1335,7 +1335,7 @@ test "the entry queue's worst case stays inside its byte budget" {
 
     // The shipped default has to be comfortably inside the budget, or the
     // out-of-the-box configuration is the one that spends it. At the widths
-    // above it costs about 17 MiB, roughly a quarter of the ceiling.
+    // above it costs under 2 MiB.
     const default_max: usize = (model.Logging{}).query_log_buffer_max;
     try testing.expect(default_max <= query_log_buffer_max);
     try testing.expect(default_max * @sizeOf(Entry) <= queue_budget_bytes / 2);