[clang][SYCL] Fix integration header for free function kernels with enum values in args

[dklochkov-emb]

This pull request fixes the problem of using enum value names with free functions.
Context of the problem:
Free function kernels use integration header to create shim functions. It is added forward declarations for this. I.e. if we have some free function kernel:

void some_kernel_func(some_args){
...
}

Wee need to emit its declaration into integration header because user sources are not "available" from header:
void some_kernel_func(some_args);

However, if any of arguments(some_args) are classes, structs etc we need to emit their forward declarations
Using enum does not produce problems but using enum value names emits build error because compiler does not allow to use forward declaration of particular enum value, i.e.

enum class A{
A0.
A1
};

it is not allowed to use forward declaration of A::A0

This pull request adds workaround solution of the problem: every enum value name is replaced with its value casted to enum type during sema analysis. I.e.
A::A1 is replaced with static_cast<A>(1);

[Fznamznon]

I'm just curious about what it will print for something like

enum class A : int {
  B,
  C
};

template <A a, typename B> struct C{

};

void freefunctionkernel(C<static_cast<A>(38), int> cc);

[dklochkov-emb]

I'm just curious about what it will print for something like

enum class A : int {
  B,
  C
};

template <A a, typename B> struct C{

};

void freefunctionkernel(C<static_cast<A>(38), int> cc);

Integration header will contain something like this:
`
enum class A : int;
template <A a, typename B> struct C;

void freefunctionkernel(C<static_cast(38), int> cc);
static constexpr auto __sycl_shim() {
return (void (*)(struct C<static_cast(38), int>))freefunctionkernel;
}
`

[Fznamznon]

static_cast(38) won't compile right? What if the value is within the range of enum values?

[dklochkov-emb]

static_cast(38) won't compile right? What if the value is within the range of enum values?

I did not manage to compile with this example both cases - within the range and out of it:
void freefunctionkernel(C<static_cast<A>(38), int> cc);

However, when I set something easier and closer to bugs:
at::native::xpu::OpSFunctor<int, static_cast<at::native::xpu::OP_MODE_SCOPED>(2)> at::native::xpu::OpSFunctor<int, static_cast<at::native::xpu::OP_MODE_SCOPED>(32)>
It is compiled without errors.