Full Go/Cobra port of sumctl to replace the Python CLI (owner-directed, supersedes #38/#39 prototype scope)

[douglasjarquin]

Decision

Owner (douglasjarquin) has decided to proceed with a full Go/Cobra port of sumctl, replacing the Python implementation as the production CLI. This is separate from and does not reopen #38/#39: those issues correctly closed a bounded, benchmark-gated prototype whose measured surface did not clear the adoption gate (see PR #90, benchmarks/issue-38/report.md). PR #90 stays open on its own merits and is not touched by this issue.

This issue authorizes what #38 explicitly said not to do by default: translate the full sumctl.py (and its lib/ modules) into Go, not just the previously measured narrow surface.

Scope

• Full command-for-command parity port of sumctl (all subcommands currently in bin/sumctl / lib/sumctl.py and its lib/ modules) to Go using Cobra, per the roadmap's Roadmap: reliability first, independently usable feature tracks, and direct Remainder integration #29 Go/Cobra decision (spf13/cobra, no Viper, no second parser).
• Exact behavior parity: command paths/names, flags (including leading-dash values, repeated options, global vs local flags, --), aliases, help text, stdout/JSON shapes, exit codes, env/settings precedence.
• Domain/state/Herdr logic must not depend on cobra.Command — small ordinary Go functions own operations, consistent with Build the Sum-owned Herdr Mesh builtin in Go with Cobra, preserving CLI and MCP contracts #41's Mesh builtin split.
• Differential testing against the frozen Python reference revision: outputs, filesystem/Git/Herdr side effects, state/brief/evidence formats, role/instance/occupant checks, locking/atomic writes, subprocess argv, cancellation/timeouts.
• Reuse the existing benchmarks/issue-38/ measurements and in-repo Go module scaffolding where still applicable; do not repeat Measure actual hot-path costs before choosing any Sum rewrite #37's measurement work merely to re-justify the language choice — this issue's authorization comes from the owner, not a new benchmark gate.
• Preserve worker verification → independent root verification → human review → human merge. No installation update, default binary switch, or fleet restart as part of this issue; that is Conditionally roll out the Go/Cobra helper surface with known-good rollback #39-equivalent rollout, tracked separately once parity is demonstrated.
• Unported edge cases must be explicitly listed rather than silently dropped.

Out of scope

• Build the Sum-owned Herdr Mesh builtin in Go with Cobra, preserving CLI and MCP contracts #41's native Go Mesh MCP builtin (independent, already Go/Cobra by its own decision).
• Any change to Prototype the measured Go helper surface with Cobra and exact behavior parity #38/Conditionally roll out the Go/Cobra helper surface with known-good rollback #39's closed status or PR refactor(helper): retire the unadopted Go prototype #90.
• Production cutover/activation (depends on safe rollout groundwork, tracked separately).

References

Roadmap: #29 (Go/Cobra implementation decision, section 3 and section 6).
Prior bounded prototype and no-go: #38, #39, PR #90.