v3: align enum fixed-array typedef names in callbacks

[antono3]

Superseded by #28902

The refreshed #28902 at b06d4abc97644a67a2e6fcf6de20936d273335db includes this PR's sizeof(Callback) regression and passes it on current master 6c359c2282. It fixes callback key conversion while retaining backed-enum typedef names, so this separate global naming change is no longer needed.

Original proposal

Summary

V3 names fixed-array types in function pointer signatures through TypeChecker.c_type(), which renders enum elements as int. The C generator used value_c_type() for the emitted fixed-array typedef name, rendering the same enum as i32. The resulting C references Array_fixed_int_2 while only Array_fixed_i32_2 is declared.

Use TypeChecker.c_type() for enum elements when naming the generated typedef. Its element storage type remains unchanged. Add a regression case that compiles and runs a callback type with [2]Combiner through V3.

Reproducer before this change

module main
enum Combiner as i32 { keep = 0 replace = 1 }
type Callback = fn (command voidptr, ops [2]Combiner)
fn main() { _ = sizeof(Callback) }

v -new-compiler -cc gcc main.v fails with unknown type name 'Array_fixed_int_2'.

Validation

• ./v self
• ./v fmt -w on both touched files
• ./v -silent test vlib/v/compiler_tests/fixed_array_typedef_codegen_test.v (pass)
• VJOBS=4 ./v -silent test vlib/v/gen/c/ (25/25 pass)
• The standalone reproducer compiles with ./v -new-compiler -cc gcc after the fix.

I also attempted ./v -silent vlib/v/compiler_errors_test.v; the current V3
runner disagreed with existing V1-format golden output for unrelated assertion
fixtures, so I stopped that broad run after the mismatch.

This was found while testing the Vulkan raw bindings' V3 C compatibility; the binding's platform declarations are separately guarded.