bug engine: don't hang on 4000+ tasks with stack usage monitor

A bug was found where creating 4000+ tasks at the same time with stack usage monitor enabled would hang the process in release builds.

As a bandaid, only enable stack usage monitor for no more than 1000 tasks. This means that some stack overflows in high-load services may have poor diagnostics.
commit_hash:f2fbcea2813caffe437adacd9edd6b7271dc64d8

Author: Anton3

core/src/engine/coro/pool.cpp

@@ -127,7 +127,9 @@ Pool::Coroutine Pool::CreateCoroutine(bool quiet) {
             LOG_DEBUG() << "Created a coroutine #" << new_total << '/' << config_.max_size;
         }
 
-        stack_usage_monitor_.Register(coroutine);
+        if (new_total <= kStackUsageMonitorLimit) {
+            stack_usage_monitor_.Register(coroutine);
+        }
 
         return coroutine;
     } catch (const std::bad_alloc&) {

core/src/engine/coro/stack_usage_monitor.hpp

@@ -40,6 +40,10 @@ class StackUsageMonitor final {
     utils::FastPimpl<Impl, 1024, 8> impl_;
 };
 
+// It was discovered experimentally that registering more than ~4000 coroutines at the same time may crash or hang
+// the service on Linux x86_64 after UFFDIO_REGISTER failure.
+constexpr std::size_t kStackUsageMonitorLimit = 1000;
+
 std::size_t GetCurrentTaskStackUsageBytes() noexcept;
 
 }  // namespace engine::coro