refactor(helper): retire the unadopted Go prototype

[douglasjarquin]

Summary

Retire the unused Go helper after its compiled benchmark failed the existing adoption gate.
Native version/help are faster, but no stateful command moved to Go, frequency-weighted benefit is unestablished, parity is unevaluated, and whole-operation memory is incomparable.
The measurements and historical reproduction remain in benchmarks/issue-38/.
Production stays on Python; the separately shipped Go Mesh and its Cobra/MCP dependencies remain available.

This takes #38's explicit discard route and records #39 as deferred/not planned.
Closes #38.
Closes #39.

Read benchmarks/issue-38/report.md for the decision, then verify_release in lib/sumctl.py for the packaging boundary.
New releases require the retained Mesh artifact; historical releases use their own bundled inventory.
Verification now binds manifest inventory to the hash-checked bundled file and requires canonical native paths, matching source/version, and the current host platform.
This establishes bundle consistency, not signature or executable-header attestation.

The PR is stacked on #87's reviewed CI baseline to reuse its fixture fixes.
No installation update, default switch, fleet restart, or installed-binary removal occurs here.
Remainder, Pinchos, and other roadmap tracks do not depend on this no-go decision.

Validation

• I followed the contributor checklist.
• I recorded the canonical verification command, result, and evidence path.
• I recorded applicable manual or live checks, including anything not run.
• I stated deployment or release impact when relevant.

Candidate: 7cdb7a41abf8f7403162c392fabddf067adced39.
Base: d9d0bd210aa06d4dd87215460a642e1a4a2979aa.
Command: MISE_ENABLE_TOOLS=go,python,node python3 .agents/skills/verify/scripts/verify_run.py --base d9d0bd210aa06d4dd87215460a642e1a4a2979aa.
Runs use existing pinned tools and isolated temporary mise state/cache, with hooks and automatic installs disabled.

Worker verification passed: 536 Python tests with three skips, 10 Node tests, and the demo.
Record: issue38-audit/.artifacts/verification/20260910T023943Z-aa79b075/run.json.
Independent root verification passed the same checks on the same clean candidate.
Record: issue38-root/.artifacts/verification/20260910T025539Z-b218846a/run.json.

All five exact-candidate review areas passed.
Actual CLI checks accepted current and archived historical releases and refused altered inventory, omitted inventory, redirected paths, and foreign-platform metadata.
Help and invalid-option controls passed; all temporary fixtures were cleaned.
The retained cgo-free Mesh binary ran with an empty PATH.
Review records: issue38-root/.artifacts/review/issue38/ledger.md and qa-7cdb7a4.md.
Final HEAVY review approved the exact candidate, including explicit review of the changed verification policy.
Report: issue38-root/.artifacts/review/issue38/heavy-7cdb7a4.md.

The aggregate reports 76 automated scenarios and nine manual scenarios not run.
The changed feature map requires independent policy review and cannot certify itself.
No authenticated Herdr canary or deployment is claimed; human merge remains required.