Make crates.io publishes retryable and resumable

[IvGolovach]

Summary

Makes the crates.io release publish chain resumable after partial uploads and resilient to crates.io new-crate rate limiting.

The publish script now checks whether each crate@workspace_version is already visible on crates.io before real publishing. Already-published crate versions are skipped, missing crate versions are published, and HTTP 429 rate-limit responses are retried with a bounded retry loop using the crates.io retry timestamp when one is provided.

Why

Issue #691 showed the v0.66.0 release could publish GitHub artifacts successfully while the crates.io chain failed partway through after several crates had already uploaded. A plain rerun was not safe enough because the script did not distinguish already-published crates from missing crates, and it treated crates.io 429 responses as permanent publish failures.

This keeps the GitHub release/tag surface separate from crates.io recovery: operators can rerun the publish chain for the same release tag, skip crate versions that already landed, and continue with the missing downstream crates.

Diff scope

• Adds crates.io registry status classification for published, missing, and unknown crate versions.
• Skips already-published crate versions before real cargo publish attempts.
• Retries only crates.io rate-limit failures, with configurable max attempts and fallback backoff.
• Parses the Please try again after ... timestamp from crates.io when available, with Linux/macOS date parsing support.
• Fails closed if the registry status cannot be verified before a real publish.
• Redacts CARGO_REGISTRY_TOKEN from captured cargo diagnostics before replaying output.
• Adds process-level script tests with mocked cargo, curl, sleep, and date for retry/resume/error behavior.
• Updates release docs with the missing mesh-llm-api-client crate and partial-publish recovery guidance.

Compatibility

Release tooling/docs only. No runtime behavior, crate API, mesh protocol, protobuf, Skippy ABI, plugin protocol, or release version change.

Prerelease behavior is unchanged: crates.io publishing remains gated to non-prerelease release tags by the existing workflow.

Branch integrity

• Base branch: main
• Validated base: origin/main@5c805ead3449c54de631ed4d60fd00ab89d57c0f
• Branch state at local validation: 0 behind / 1 ahead
• Head commit: 602d21e3908eeafcc62297f0bf1ab1404ca14e71

Commit integrity

One logical commit:

• 602d21e3908eeafcc62297f0bf1ab1404ca14e71 - Harden crates.io publish retries

Final diff contains only release docs, the crates.io publish script, and targeted script tests.

Validation

• Validation tier: Tier 4 - release publish verification tooling and crates.io partial-publish recovery for issue Release crates.io publish chain breaks on HTTP 429 (new-crate rate limit) #691.
• git fetch --no-tags origin main:refs/remotes/origin/main: PASS, origin/main at 5c805ead3449c54de631ed4d60fd00ab89d57c0f.
• git diff --check: PASS, no output
• git diff --cached --check: PASS, no output
• bash -n scripts/publish-crates.sh: PASS
• scripts/publish-crates.sh --allow-dirty: PASS, rejected because --allow-dirty requires --dry-run.
• python3 -m unittest scripts.tests.test_publish_crates: PASS, 7 passed
• python3 -m unittest discover -s scripts/tests: PASS, 27 passed
• cargo run -p xtask -- repo-consistency release-targets: PASS
• CARGO_TARGET_DIR=$(mktemp -d) scripts/publish-crates.sh --dry-run --allow-dirty: PASS; dry-run verified publishable crates through model-artifact and deferred downstream crates whose 0.66.0 registry dependencies are not published yet.
• Ledger: not applicable - not required for selected validation tier/change family.
• Version: not applicable - release publish tooling/docs only; no release version sync required.
• Not run: actionlint - not installed locally and workflow file was not changed.
• Not run: real cargo publish / just release - intentionally avoided for PR validation.

Runtime safety

Not applicable - no runtime path changed.

CI context confirmation

No workflow file or CI context names changed. Required PR CI will run after PR creation/update.

Rollback

Rollback: revert this PR.

DB downgrade: not applicable.
Data repair: not applicable.
Operational caveats: none known.

Known residual risks

This PR cannot prove a real crates.io upload in PR validation because it intentionally does not use a registry token or publish crates. The first production rerun should verify all 11 crate versions on crates.io after the release publish job completes.

crates.io can still enforce account-level new-crate rate limits; this PR makes the release job resumable and retry-aware, but it does not replace requesting a crates.io rate-limit increase if the publish chain continues to grow.

[i386]

LGTM