Improve test coverage for actions

[volsa]

(Disclaimer: Generated with Claude, based on the discussion here #1545 (comment); proof-read by me)

---

Summary

A recent fix (2b3388f) indirectly addressed a segfault when calling actions with inherited VAR_IN_OUT parameters, but no existing tests covered this case.

Example

// cargo r -- target/demo.st tests/lit/util/printf.pli --linker=clang && ./demo.st.out
PROGRAM mainProg
    VAR
        fb : fb_t;
        localBool: BOOL;
    END_VAR

    localBool := TRUE;
    fb.foo(myInOut := localBool);

    localBool := FALSE;
    fb.foo(myInOut := localBool);
END_PROGRAM

FUNCTION_BLOCK fb_t
    VAR_IN_OUT
        myInOut : BOOL;
    END_VAR
END_FUNCTION_BLOCK

ACTIONS
    ACTION foo
        printf('myInout = %d$N', myInOut);
    END_ACTION
END_ACTIONS

FUNCTION main
    mainProg();
END_FUNCTION

Cause

The generated IR only loaded localBool's value but never stored its address into fb.myInOut, leaving the pointer NULL:

  %fb_t = type { ptr, ptr }

  ; Before (broken)
  %load_localBool = load i8, ptr %localBool, align 1  ; loaded but never used
  call void @fb_t__foo(ptr %fb)                        ; myInOut still NULL → segfault

  ; After (fixed)
  %1 = getelementptr inbounds %fb_t, ptr %fb, i32 0, i32 1  ; get myInOut field
  store ptr %localBool, ptr %1, align 8                     ; store address
  call void @fb_t__foo(ptr %fb)                             ; myInOut now valid

We should add test coverage to prevent regressions.