ci: add the org-wide reusable workflow alongside the existing pipeline

[JustinKovacich]

What

Adds .github/workflows/main.yml, a thin caller for luminartech/rust_workflow@v1, matching simple_doip, uds_protocol and automotive_wire_codec. Deletes release-plz.yml — the shared workflow runs the same two release-plz jobs behind use-release-plz: true — aligns release-plz.toml with the siblings, and adds the deny.toml the security job needs.

It takes over none of this repo's eight existing jobs. I opened this PR expecting two to move; CI disproved that, and the description below is the corrected version. ci.yml survives intact, renamed to Bare-metal & Platform CI.

What the shared workflow actually contributes

All of these are new here, not replacements:

new job
Release-plz PR / Release	replaces this repo's own release-plz.yml
Documentation	rustdoc with warnings denied
Security Audit	cargo-audit + cargo-deny
Publish Dry Run	proves the crate still packages
Build	release build
Build no_std (bare-metal)	smoke test alongside the local gate

The release plumbing is the real win. Versioning, changelog, tags, GitHub releases and the crates.io publish now run through the same jobs as the other three protocol crates, off a release-plz.toml whose [changelog] parsers are byte-identical to theirs.

Why every existing job stayed

Each of these was tried or analysed, not assumed:

• Build, Test & Coverage — the shared workflow applies one feature-flags value to build, test, clippy and docs. This crate needs two: lint wants bare_metal in, tests want it out, because the server's runtime caps default tight under bare_metal and generous otherwise. Measured: the host suite with bare_metal on fails 12 server tests; without it the library is 563 passed, 0 failed.
• Format & Lint — the shared lint job hardcodes cargo clippy --all-targets, which this pipeline never used. That surfaces ~89 pre-existing pedantic and style findings in test and example code. clippy-args only appends after --, so the scope cannot be narrowed from the caller.
• SemVer Check — cargo-semver-checks builds rustdoc with all features and takes no feature input, so it hits this crate's own guard: feature `bare-metal-runtime` is no-alloc and cannot be combined with the alloc features. Not fixable from the caller. The local job passes a feature set.
• Build & Test (Windows) — every job in rust-ci.yml is runs-on: ubuntu-latest, and its header says that is intentional.
• no_std target build — builds four bare-metal feature combinations separately and audits each rlib for allocator symbols. run-no-std builds one target with and without the alloc feature. Both run now: the shared one as a smoke test, this as the gate.
• build-std core gate — certifies against a core-only sysroot. No equivalent input.
• Bare-metal runtime (nightly) — nightly-only and mutually exclusive with the alloc features; the shared workflow's nightly is used only for fuzz and miri.
• Linear PR History — not provided.

Stages deliberately off

input	why
run-unit-tests	two feature regimes, one feature-flags knob
run-lint	--all-targets scope, see above
run-semver-checks	all-features rustdoc build, see above
run-property-tests	no prop_-named tests; cargo nextest exits 4 on an empty selection, so it would fail, not skip
run-fuzz-tests	no fuzz/ package
run-pre-commit	no .pre-commit-config.yaml
run-msrv	no rust-version in the manifest — a policy call, not a CI one
run-miri-tests	never run here; enabling a new gate inside a pipeline migration makes any failure ambiguous

Also: --all-features cannot build this crate at all, so feature-flags is set to the old $ALLOC_FEATURES, and alloc-feature to _alloc (the workflow default is alloc; naming it wrong would silently skip the alloc build variants).

deny.toml

The one CI failure that was missing scaffolding rather than a workflow mismatch. cargo-deny's default license allowlist is empty, so with no config it rejected every dependency — including MIT OR Apache-2.0 — with "license is not explicitly allowed". Copied from simple_doip.

No pr-lint.yml, unlike simple_doip

simple_doip has one because it squash-merges with squash_merge_commit_title: PR_TITLE, so the PR title is the commit subject release-plz parses. This repo has allow_squash_merge: false and merges with merge commits, so the branch's own commit subjects are what release-plz reads. A PR-title lint would give false assurance while leaving the text that actually drives versioning unchecked.

Follow-up worth considering: nothing currently enforces conventional commits on individual commits here, and those are what drive versioning. Either add a commit-lint, or switch the repo to squash-merge like the siblings.

The general finding

Every blocker traces to one assumption in rust_workflow: that --all-features builds, and that a single feature set suits build, test, clippy, docs and semver-checks alike. That holds for the three sibling crates. It cannot hold for a crate with mutually exclusive no-alloc and alloc feature sets.

Three inputs would let this crate actually converge — a test-specific feature set, a lint-scope toggle, and a features input for semver-checks. That is a change to the shared workflow affecting every repo, so it is deliberately not in this PR.

Verification

CI on this branch: 13 pass, 0 failures. Both workflows appear, no job from the previous set is lost, and every disabled stage reports skipping rather than silently vanishing.

🤖 Generated with Claude Code

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

@@           Coverage Diff           @@
##             main     #156   +/-   ##
=======================================
  Coverage   81.72%   81.72%           
=======================================
  Files          48       48           
  Lines       16178    16178           
=======================================
  Hits        13222    13222           
  Misses       2956     2956