Fix access to the DWT peripheral for STM32 HAL (#12434)

Access to the DWT peripheral for the `CYCCNT` register needs to happen before `main()`. The code needs to be called after the setup of the system clocks, so the right place is between the `premain()` and `main()` function of the STM32 Arduino core.

This patch moves the DWT access code to a new function, which is then placed between `premain()` and `main()`.

Marlin/src/HAL/HAL_STM32/HAL.cpp

@@ -78,14 +78,20 @@ uint16_t HAL_adc_result;
 // Public functions
 // --------------------------------------------------------------------------
 
-// HAL initialization task
-void HAL_init(void) {
 
-  // Needed for DELAY_NS() / DELAY_US() on CORTEX-M7
-  #if (defined(__arm__) || defined(__thumb__)) && __CORTEX_M == 7
+// Needed for DELAY_NS() / DELAY_US() on CORTEX-M7
+#if (defined(__arm__) || defined(__thumb__)) && __CORTEX_M == 7
+  // HAL pre-initialization task
+  // Force the preinit function to run between the premain() and main() function
+  // of the STM32 arduino core
+  __attribute__((constructor (102)))
+  void HAL_preinit() {
     enableCycleCounter();
-  #endif
+  }
+#endif
 
+// HAL initialization task
+void HAL_init(void) {
   FastIO_init();
 
   #if ENABLED(SDSUPPORT)