[validation-only] keep named kubectl get unfiltered (#4613)

[1932819331hy-debug]

Fork validation counterpart for upstream rtk-ai#4667 (issue rtk-ai#4613).

Current head: 5e58a70a662740a57488d8ffe1a4eb75753efaee. The latest update repairs only generated commit author/committer metadata so GitHub associates every commit with the account. No source files, commit messages, or author dates changed. Source tree 906064b9644eaff258c7af9ff2fdb3d262ea2326 is byte-identical to the tested prior head 22731a9e7c0127e0a18dacec6268df7dbc5f2e11.

Named pod/service queries preserve kubectl's native output, original arguments, stderr and exit status. The implementation uses the shared argument tokenizer, retains compact recognized list queries, and passes output requests/unknown/incomplete options to kubectl. Only the container routing source and five process regression tests differ from upstream develop 4ba5ac0.

Local checks executed on prior head 22731a9 (same source tree), Rust 1.91.0 and real Go 1.26.9 on PATH:

• cargo +1.91.0 fmt --all --check and proxy workspace format check passed.
• cargo +1.91.0 clippy --workspace --all-targets --locked passed with no warnings.
• cargo +1.91.0 test --all --locked: 5,000 passed, 0 failed, 8 ignored.
• On unmodified upstream source e542dca, four of five process regressions fail and the list control passes; all five pass with the named-query fix.
• Two argument regression groups fail on the handwritten scanner and pass with the tokenizer. Independent source review found no material issue.

No broad local tests were repeated for the identity-only update. Exact-new-head fork CI now PASS on 5e58a70a662740a57488d8ffe1a4eb75753efaee: run 39319925049, attempt 2. The previous attempt's macOS job was cancelled, while the other jobs passed. On 2026-10-20 the macOS test matrix job alone was rerun successfully; the complete workflow now reports success. All 14 jobs are successful at the latest attempt, including fmt, Clippy, rustdoc, proxy lint/tests, benchmark, Ubuntu, macOS, Windows (both toolchains), Semgrep, security and OSS purity. The remaining evidence gap is a real Kubernetes cluster and upstream CI/maintainer acceptance. Upstream CLA passes. No CLA signing or legal acceptance was performed by the agent.

Fork develop was synchronized by a preserving-history merge that retains all six preexisting probe commits and has the exact upstream source tree. The alternative upstream PR rtk-ai#4623 normalizes single-object JSON rather than selecting native named-query output; maintainers still choose their preferred approach.

Implementation and review were AI-assisted. Process tests use fake kubectl and the existing pod fixture, not a real cluster. Exact-head fork CI passes, but upstream CI and maintainer review remain pending.