std::mem::size_of::<Result<char,u16>>() says 8 when 4 should be feasible

I tried this code:

fn main() {
    println!("{}", std::mem::size_of::<Result<char,u16>>());
}

I expected to see this happen:

Expected it to print 4. Expected the Ok case to have the same representation as char. Expected the Err case to have a non-zero discriminant value in the upper 11 bits (upper 11 bits for char are always zero) and expected the u16 to reside in the two least significant bytes.

Instead, this happened:
Got 8, i.e. the compiler didn't make good use the top 11 bits of 32 bits that u16 doesn't occupy at all and that are zero for char, since char only uses the lower 21 bits.

There should be enough available discriminant bits that Option<Result<char,u16>> could have size 4, too.

Use case: Surfacing the unpaired surrogate as Err when decoding UTF-16.

Meta

Tried the current stable and nightly on https://play.rust-lang.org/?version=nightly&mode=release&edition=2021

It said:
Stable channel

Build using the Stable version: 1.74.0
Beta channel

Build using the Beta version: 1.75.0-beta.3

(2023-11-20 b66b7951b9b4258fc433)
Nightly channel

Build using the Nightly version: 1.76.0-nightly

(2023-11-26 6cf088810f66fff15d05)

Reproduce the reported layout with the provided std::mem::size_of::<Result<char,u16>>() example on the Rust Playground or a local nightly compiler. Investigate Rust's enum layout and niche optimization behavior; done means the requested Result and Option<Result<char,u16>> representations use 4 bytes without breaking layout guarantees.