Confusing lifetime error when using HRTB and closure

I tried this code:

trait StorageIterator {
    type KeyType<'a> where Self: 'a;
}

fn f<I: for<'a> StorageIterator<KeyType<'a> = &'a [u8]>>() {
    || true;
}

I expected to see this happen: This code can pass compilation.

Instead, this happened: Rustc reports a lifetime error.

error: `I` does not live long enough
 --> <source>:6:5
  |
6 |     || true;
  |     ^^^^^^^

error: aborting due to 1 previous error

Following situations can pass compilation:

1. remove closure

trait StorageIterator {
    type KeyType<'a> where Self: 'a;
}

fn f<I: for<'a> StorageIterator<KeyType<'a> = &'a [u8]>>() {}

2. remove where bound in trait

trait StorageIterator {
    type KeyType<'a>;
}

fn f<I: for<'a> StorageIterator<KeyType<'a> = &'a [u8]>>() {
    || true;
}

3. add 'static for generic I

trait StorageIterator {
    type KeyType<'a> where Self: 'a;
}

fn f<I: for<'a> StorageIterator<KeyType<'a> = &'a [u8]> + 'static>() {
    || true;
}

Meta

rustc --version --verbose:

rustc 1.88.0 (6b00bc388 2025-06-23)
binary: rustc
commit-hash: 6b00bc3880198600130e1cf62b8f8a93494488cc
commit-date: 2025-06-23
host: x86_64-unknown-linux-gnu
release: 1.88.0
LLVM version: 20.1.5

Start by reproducing the minimal example with rustc 1.88.0 and compare it with the three compiling variants in the issue. Investigate how HRTBs, generic associated types, the where Self: 'a bound, and closure creation interact. Done means the example compiles as expected or the diagnostic accurately explains the required lifetime constraint, with regression coverage added in the relevant compiler tests.