workspace,scripts: mark workspace-only crates publish = false and enforce it when publishing

[avi-starkware]

Only the 16 crates with a version in the root workspace manifest are ever published; the other 99 members (97 path-only workspace deps plus toml_test_utils and workspace_tests) are referenced by path alone. This PR states that explicitly with publish = false, which makes cargo publish refuse them and lets cargo-deny's allow-wildcard-paths treat their versionless path dependencies as intended rather than as wildcards (next PR).

scripts/publish_crates.py

• Fails early if the publish order contains a publish = false crate (via cargo metadata; no new Python dependency).
• Follows only normal,build edges in cargo tree. Its default edges include the root crate's dev-dependencies, so publishing apollo_config would have attempted apollo_test_utils, and blockifier would have pulled in blockifier_test_utils.

Test: workspace_tests::test_members_are_unpublishable_iff_they_are_not_for_publish asserts publish = false ⟺ not in the version-derived publish set, so the flag and the version rule cannot drift apart (verified to fail when a flag is removed).

Verified: cargo metadata → 99 restricted / 16 unrestricted, and the 16 are exactly the versioned set; taplo format --check clean; cargo check --workspace --all-targets green; publish orders for blockifier, starknet_api, apollo_starknet_os_program, blockifier_test_utils are all-publishable with no dev-dep leakage.

Deliberately not touched: starknet_proof_verifier is versioned (publishable) but depends on the git dependency privacy-circuit-verify-v2 with no crates.io version, so it cannot be published today. cargo-deny keeps reporting it as one wildcard finding; that is accurate and intentionally left visible.

🤖 Generated with Claude Code

[reviewable-StarkWare]

This change is 

[github-actions]

Artifacts upload workflows:

• Started at 2026-09-14T09:51:15Z
• Started at 2026-09-14T09:52:57Z
• Started at 2026-09-14T10:37:47Z
• Started at 2026-09-14T10:39:01Z

[cursor]

PR Summary

Low Risk
Manifest and tooling-only changes; no runtime or published-crate behavior changes beyond explicit publish restrictions.

Overview
Marks 99 workspace-only members (sequencer internals, test helpers, CLIs, etc.) with publish = false so cargo publish rejects them and tooling can treat them as private. The 16 crates.io-bound packages (those with a version in the root manifest) are unchanged and remain publishable.

scripts/publish_crates.py now builds the dependency order from normal,build edges only, so dev-only crates like apollo_test_utils / blockifier_test_utils are not pulled into publish runs. It also reads cargo metadata to assert the publish chain contains no publish = false members before attempting uploads.

workspace_tests adds test_members_are_unpublishable_iff_they_are_not_for_publish, keeping publish = false in sync with the existing “version iff for publish” rule, with a short doc cross-link in the version test.

^{Reviewed by Cursor Bugbot for commit 7d9a8fa. Bugbot is set up for automated code reviews on this repo. Configure here.}

[dorimedini-starkware]

@dorimedini-starkware reviewed 1 file.
Reviewable status: 1 of 101 files reviewed, all discussions resolved (waiting on asaf-sw).

[einat-starkware]

@einat-starkware reviewed 101 files and all commit messages, and made 1 comment.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on asaf-sw).