chore: prepare strided-rs v0.4.0

[shinaoka]

Summary

• prepare all nine strided-rs workspace crates for 0.4.0
• retain strided-einsum2, strided-opteinsum, and both adapters as maintained CPU einsum components
• add targeted crates.io keywords/categories and accurate per-package license/provenance files
• update release docs for tagged, dependency-ordered, resumable crates.io publication

This supersedes the retirement direction in #199 for the four einsum crates.

Verification

• cargo fmt --all -- --check
• cargo test --workspace (898 passed, 9 ignored)
• RUSTDOCFLAGS='-D warnings' cargo doc --workspace --no-deps
• feature matrix for facade, parallel kernel, and permutation paths
• cargo llvm-cov --workspace --json --output-path coverage.json
• python3 scripts/check-coverage.py coverage.json (54/54 files)
• python3 scripts/test-repository-rules-review.py (79 passed)
• deterministic and DeepSeek repository-rules reviews (zero findings)
• all nine package lists and metadata inspected
• cargo publish -p strided-traits --dry-run
• crates.io ownership and 0.4.0 absence verified for all nine crates

Strict Rust 1.97 clippy with -D warnings reports pre-existing findings identically on origin/main; this PR changes no Rust source. Full workspace clippy was also run with --cap-lints warn to inventory the baseline without expanding release scope.

Release sequencing

After merge and exact-SHA CI success, tag v0.4.0; package, inspect, dry-run, publish, and wait for registry visibility one crate at a time in dependency order. No manifest changes occur after tagging.

[github-actions]

Repository rules review

Repository rules review (0082be119828af37da535c1cf5ba3c9b91092a1c...6e586b26baebda81bebdff9c109b53616311a184)
Verdict: pass
LLM review: 4 chunk(s) (56471 chars, 59942 chars, 49079 chars, 15035 chars) in 74.1s; 0 finding(s) returned, 0 kept, 0 dropped by diff-anchor filtering.
No findings.