balance synthetic add and cancel rates so the book is steady-state under any seed

PR #24 stabilized the book under seed 11 but the live deploy uses seed
42, which (as a property of that particular RNG path) produces a
much less mobile reference and lets both sides accumulate to 1500+
qty even with the cancel-window targeting in place. Verified locally
with seed 42 plus the rate=400, 72/8/20 mix: top size sat at 1207 to
3087 after a 25-second warm-up. That is the right fix in the wrong
place: the cancel window helps cancels actually hit resting orders,
but the per-event cancel share is still too low to balance the per-
event add share.

Shift the mix from 72 / 8 / 20 to 50 / 10 / 40 (limits / markets /
cancels). At qty range [1, 6], the expected per-second add qty
(~700) now matches the expected per-second removal qty from cancels
plus market sweeps (~640), so the book reaches a steady state in
the low hundreds rather than accumulating. The 40 percent cancel
share is still a visually legible undercount of what real-exchange
ITCH feeds actually show (often 80 percent plus cancels), but it
reads as a busy market.

Local verification with seed 42 (the live deploy's seed), 30s warm-up
plus 20s sample at rate 400:

  top size min / median / max = 277 / 338 / 728

Compared to the previous run on the same seed and same code minus
this commit, which sat at 1207 to 3087. 182/182 release tests still
pass.

Author: MustafaNazeer

apps/server/main.cpp

@@ -266,8 +266,16 @@ int main(int argc, char** argv) {
             if (reference < kPriceFloor) reference = kPriceFloor;
             if (reference > kPriceCeiling) reference = kPriceCeiling;
 
+            // Event mix tuned for steady state: at this share with the
+            // current qty range, the expected add and cancel rates are
+            // roughly balanced, so the resting book hovers in the low
+            // hundreds rather than accumulating into the thousands.
+            // Real-exchange ITCH feeds show much higher cancel ratios
+            // than this (often 80 percent plus); 50 / 10 / 40 is a
+            // visually legible undercount that still reads as a busy
+            // market.
             const int b = bucket(rng);
-            if (b < 72 || max_id == 0) {
+            if (b < 50 || max_id == 0) {
                 // Limit order placed near the reference, with a small
                 // offset so the resting book builds depth across a
                 // handful of ticks rather than at one price.
@@ -280,7 +288,7 @@ int main(int argc, char** argv) {
                     : reference + offset;
                 ev.qty = qty_dist(rng);
                 max_id = ev.order_id;
-            } else if (b < 80) {
+            } else if (b < 60) {
                 // Occasional market sweep that lifts the resting book.
                 ev.kind = meridian::EventKind::NewOrder;
                 ev.type = meridian::OrderType::Market;