fix(core): Respect logical length in Buffer vectored views

[codeAnqiang-ma]

Which issue does this PR close?

Closes #8063.

Rationale for this change

For a non-contiguous Buffer, to_io_slice() and the bytes::Buf::chunks_vectored() impl destructure Inner::NonContiguous with .., dropping the logical size field, and then walk each part to its physical end. Because truncate / slice / split_to / split_off shrink size without dropping parts, both methods then hand back more bytes than remaining() — for a 2-byte view of ["abc", "def"] they expose all 6 bytes.

That violates the documented bytes::Buf::chunks_vectored contract ("the sum of the lengths of all the buffers written to dst will be less than or equal to Buf::remaining()"), and it contradicts chunk() and Iterator::next in the same type, which both clamp with .min(*size) to honor the Inner::NonContiguous invariant ("the logic view … spans size bytes").

#4481 already fixed the offset half of this problem in these same two methods; the size clamp was not added then.

What changes are included in this PR?

Both non-contiguous branches now bind size, track it as a running remaining, clamp each slice with .min(remaining), and stop once it reaches zero — mirroring what chunk() already does. Nothing else changes.

Regression tests test_vectored_views_after_slice, test_vectored_views_after_split_to and test_vectored_views_after_split_off assert that both vectored views yield exactly the logical content. They fail on main and pass here.

Test evidence

Before the fix (tests applied to unmodified main), all three fail — the vectored views expose the whole physical buffer:

---- types::buffer::tests::test_vectored_views_after_slice stdout ----
assertion failed: `(left == right)`
 (
<    b"abcdef",
<    b"abcdef",
>    b"ab",
>    b"ab",
 )

failures:
    types::buffer::tests::test_vectored_views_after_slice
    types::buffer::tests::test_vectored_views_after_split_off
    types::buffer::tests::test_vectored_views_after_split_to

test result: FAILED. 0 passed; 3 failed

After the fix:

$ cargo test -p opendal-core --lib --features "tokio/time"
test result: ok. 182 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

$ cargo fmt --all -- --check                                              # clean
$ cargo clippy -p opendal-core --all-targets --features "tokio/time" -- -D warnings   # clean

--features "tokio/time" is only needed to compile opendal-core's #[cfg(test)] code when the package is tested in isolation; it is unrelated to this change. The full-workspace --all-features clippy/nextest gate and the behavior tests were not run locally.

Are there any user-facing changes?

Yes — a bug fix; no signature or API change. A shortened non-contiguous Buffer passed to a vectored write now yields only its logical bytes instead of also emitting the truncated tail.

To be precise about the blast radius: there is no active data corruption in-tree. The only in-tree vectored write backend, compfs, takes its bytes via Buffer::by_ref().collect() — the Iterator impl, which already respects size — so it never reached this path. The defect matters at the public API and trait-contract level: to_io_slice is pub and documented for vectored writes, and chunks_vectored is a bytes::Buf impl, so downstream users and any future in-tree consumer doing a vectored write over a shortened view are affected.

AI Usage Statement

Prepared with AI assistance: Cursor, using Anthropic Claude Opus 5. The AI drafted the patch, the regression tests and this description; I reproduced the defect locally, verified the tests fail before the change and pass after it, and reviewed every line of the diff.