tests: adapt closure-capture-macro-introduced fanout to thread-limit

The 10-future sub-test in closure-capture-macro-introduced trips
MTH001 on GHA runners (3-4 CPUs) because the default host
thread_limit equals cpu_count and the test spawned 10 simultaneous
futures unconditionally.

The invariant under test is "N macro-introduced closures each
capture their own per-iteration i" -- N is a parameter, not the
property. Any N >= 2 exercises the closure-capture fix this
regression test guards. Adapt N to (- (mino-thread-limit) 1)
clamped to [2, 10] so the test scales to runner capacity and
keeps its bug-catching intent on every host.

Same shape as the T9 probe fix in mino-tests v0.8.1. Local suite
green: 1274 tests, 4557 assertions, 0 failed. release-gate 17/17.

Author: leifericf

CHANGELOG.md

@@ -1,5 +1,42 @@
 # Changelog
 
+## v0.255.12 — Fix: Adapt closure-capture-macro-introduced fanout to thread-limit
+
+With the v0.255.11 pthread_join deadlock cleared, CI got past the
+`transient-survives-gc-yield` hang and reached the next latent issue:
+`closure-capture-macro-introduced` in `tests/bc_closure_test.clj`
+spawns 10 simultaneous futures unconditionally. The default host
+`thread_limit` is `cpu_count`, and GitHub-hosted runners are 3-4 CPUs
+(macos-14, ubuntu-24.04, ubuntu-24.04-arm), so the 10-way fanout
+trips `MTH001` (thread-limit-exceeded) before the test can verify
+the closure-capture invariant.
+
+The invariant under test is "N macro-introduced closures each capture
+their own per-iteration `i`" — and N is a parameter, not the property
+under test. Any N ≥ 2 exercises the bug fix this regression test
+guards. The fix adapts the fanout to `(- (mino-thread-limit) 1)`,
+clamped to `[2, 10]`, so:
+
+- Local dev (12+ CPUs): N=10, same as before.
+- GHA macos-14 / ubuntu-24.04 (3-4 CPUs): N=3, the test runs and
+  asserts.
+- Embedder with `mino_set_thread_limit(S, 1)`: N=2, still meaningful.
+
+Same shape as the T9 probe fix in `mino-tests` v0.8.1. The local
+suite reports 1274 tests, 4557 assertions, 0 failed on this branch.
+
+### Verification
+
+- `./mino tests/run.clj` green locally (1274 tests, 4557 assertions).
+- `release-gate` 17/17 green.
+- CI matrix expected green on the v0.255.12 push (the original hang
+  cleared in v0.255.11; the thread-limit failure cleared here).
+
+### Files
+
+- `tests/bc_closure_test.clj` — adapt N to host thread limit.
+- `src/mino.h` — `MINO_VERSION_PATCH` bumped to 12.
+
 ## v0.255.11 — Fix: Yield state_lock Around pthread_join in Future Sweep
 
 The actual cause of the v0.255.x CI hang, pinpointed by v0.255.10's

src/mino.h

@@ -28,7 +28,7 @@
  */
 #define MINO_VERSION_MAJOR 0
 #define MINO_VERSION_MINOR 255
-#define MINO_VERSION_PATCH 11
+#define MINO_VERSION_PATCH 12
 
 /*
  * Human-readable version string of the *linked* runtime, e.g. "0.48.0".

tests/bc_closure_test.clj

@@ -150,11 +150,18 @@
           (future (deliver p (* i i)))
           (recur (inc i))))
       (is (= 0 @p))))
-  (testing "10 futures x 10 promises via dotimes deliver to distinct slots"
-    (let [ps (vec (repeatedly 10 promise))]
-      (dotimes [i 10]
+  (testing "N futures x N promises via dotimes deliver to distinct slots"
+    ;; The closure-capture invariant doesn't depend on N; any N >= 2
+    ;; that fires distinct futures with distinct captured i exercises
+    ;; the macro-introduced-closure path. Adapt N to the host thread
+    ;; grant so 3-4 CPU CI runners don't hit MTH001 before the test
+    ;; can verify the invariant. Clamp [2, 10].
+    (let [n  (max 2 (min 10 (- (mino-thread-limit) 1)))
+          ps (vec (repeatedly n promise))]
+      (dotimes [i n]
         (future (deliver (nth ps i) (* i i))))
-      (is (= [0 1 4 9 16 25 36 49 64 81] (mapv deref ps)))))
+      (is (= (mapv (fn [i] (* i i)) (range n))
+             (mapv deref ps)))))
   (testing "delay inside self-tail-call captures per-iteration param"
     (let [cls (atom [])]
       (defn G-delay [i]