Adopt preserve_none attribute for MUST_TAIL optimization

Despite using __attribute__((musttail)), the compiler may still emit
full stack frames and spill callee-saved registers in complex handlers
(e.g., do_lw), undermining the benefits of the threaded interpreter.

Integrate Clang's __attribute__((preserve_none)) (available since
Clang 19.1.0) into the MUST_TAIL macro. This attribute modifies the
calling convention to avoid preserving registers, thereby minimizing
stack setup and register pressure during instruction dispatch.

This ensures more consistent tail call optimization and reduces overhead
in the interpreter loop.

Closes: #601

Author: visitorckw

src/common.h

@@ -191,7 +191,11 @@ static inline uint8_t ilog2(uint32_t x)
  * even without optimizations enabled.
  */
 #if defined(__has_attribute) && __has_attribute(musttail)
+#if __has_attribute(preserve_none)
+#define MUST_TAIL __attribute__((musttail)) __attribute__((preserve_none))
+#else
 #define MUST_TAIL __attribute__((musttail))
+#endif
 #else
 #define MUST_TAIL
 #endif