Add targeted rust full CI test workflow

[starr-openai]

Why

rust-ci-full failures are expensive to debug when we need to rerun one suspect test on one CI-only platform.

What

• Add a separate manual rust-ci-full targeted test workflow.
• Let the caller select one existing full-ci nextest platform and one cargo-nextest filterset.
• Reuse the existing archive-backed per-platform nextest workflow.
• Require targeted filtersets to resolve to exactly one test before running it.

Verification

• YAML parsed with yq.
• Touched workflows passed actionlint.
• Dedicated workflow smoke passed: https://github.com/openai/codex/actions/runs/26641815341