Avoid over-wrapping optional callback parameters

[kennykerr]

When a parameter is both optional and a callback, it should not be wrapped in an Option.

Fixes: #3437

I have been thinking about removing the Option from the callback type itself. At that point we'd not need this workaround and we'd be able to distinguish between optional and required callback parameters, but this is a small and targeted fix to the immediate problem for the time being.

[riverar]

Looks OK in my testing.

One thing I observed--maybe known/intentional--is that delegate types are emitted with a Option wrapper, despite being not optional in all dependent functions. Should we drop the Optional at that site instead and then keep the existing code?

Edit: Ah you addressed this in your edit. 👍

// Metadata
unsafe HRESULT setPfnMiniPDBErrorCallback2(
  [In] void* pvContext,
  [In] PFNMINIPDBERRORCALLBACK2 pfn);

pub type PFNMINIPDBERRORCALLBACK2 = Option<
    unsafe extern "system" fn(
        pvcontext: *mut core::ffi::c_void,
        dwerrorcode: u32,
        szobjorpdb: windows_core::PCWSTR,
        szlib: windows_core::PCWSTR,
    ) -> windows_core::HRESULT,
>;

[kennykerr]

Right, I'm leaning toward making the C++ delegate type omit Option as I think that would make it easier to use those types generally and they would behave more consistently with other non-nullable types in windows-rs like interfaces that you have to explicitly wrap in Option as needed. But that's a bigger change I'll work separately.