docs: refresh oxmera-core status

[nightcityblade]

Fixes #69

Replaces the obsolete exercise-seam description with the current role of oxmera-core and records ADR-0005 as accepted now that its publishing policy has been implemented.

Validation: cargo fmt check and oxmera-core doc tests pass; repository crate docs contain no stale todo!() or exercises/ references.

[vyncint]

Approved, and merging as is. Nothing to fix.

This one was cheap to verify and I checked all of it:

• grep -rn 'todo!()' crates/*/src/lib.rs and the same for exercises/ both come back empty, which is exactly the "done when" the issue asked for.
• The replacement paragraph is accurate rather than merely non-stale: oxmera-core really is dtype, shape, strides, layout, device handles and the error taxonomy, and nothing in it knows about tensors or dispatch.
• RUSTDOCFLAGS="-D warnings" cargo doc -p oxmera-core --no-deps is clean, so the new text is what docs.rs will show for the next release.
• ADR-0005 → Accepted is right, and its own postscript already recorded the completion on 2026-08-21: the ten crates published, trusted publishers configured, the token revoked. The status line was the only thing still saying otherwise.

One note on mechanics, which is about this repository and not about your PR. I am merging this with rebase rather than squash. GitHub authors a squash commit from the merging account's address rather than the one on your commits, and this repository's DCO check insists the sign-off match the author exactly. Your commit already carries Signed-off-by: nightcityblade <nightcityblade@gmail.com> against that same author, so replaying it preserves a commit that passes on its own terms. The check needs widening, which is mine to do; you did nothing wrong.

Thank you for these three. Precise, self-contained, and each one closed the issue it claimed to.