fix(table output): sort tags so table output is deterministic

[dangrondahl]

The flow and environment table printers rendered tags by iterating the tags map directly, so the order of the rendered pairs changed between runs.

Against a flow tagged app=api, env=prod, team=platform, the released CLI returned three distinct orderings across 20 runs:

$ for i in $(seq 20); do kosli get flow tag-demo --org <org> | grep '^Tags:'; done | sort -u
Tags:                [app=api], [env=prod], [team=platform]
Tags:                [env=prod], [team=platform], [app=api]
Tags:                [team=platform], [app=api], [env=prod]

With this change the same loop returns a single line.

What changed

Tag rendering goes through one sorted helper in cmd/kosli/tableHelpers.go, shared by the six places that each had their own copy: get flow, get environment, list flows, list environments, get repo and get control (list repos shares get repo's).

Both existing output formats are preserved, so ordering is the only user-visible change:

• formatTags — bracketed pairs ([app=api], [env=prod]) for the flow and environment tables
• sortedTagPairs — unbracketed pairs (app=api, env=prod) for repo and control

The helper additionally guards a missing or non-map tags value, which get environment asserted unchecked, and formats values with %v rather than %s. Neither is reachable through the API as it behaves today — it sends "tags": {} for an untagged resource, and tags set via kosli tag are always strings — but the sibling printers already guarded this way, so the shared helper does too.

Every existing assertion on tag output used a single tag or none, which is why the ordering was never caught. The new tests use three keys.

Verification

• go build ./..., make vet, make lint (0 issues)
• go test ./cmd/kosli/ -run 'TestSortedTagPairs|TestFormatTags|TestTagRendering|TestPrintEnvironmentAsTableWithoutTags'
• Manually against app.kosli.com, released binary vs. this build, as above

Not yet run: the command suites (GetFlow, ListFlows, GetEnvironment, EnvironmentLs, GetRepo, GetControl, ListControls). GetRepo and GetControl have golden assertions on tag output, so these want a run before merge.

Follow-ups (not in this PR)

Two more clusters of duplication in the table printers:

• Pagination footer — 5 sites printing the same line from 3 different response envelope shapes, with the noun drifting (items / controls / repos)
• Empty-result message — 17 sites hand-rolling "No X were found", 8 with the page-number variant

Deciding whether [k=v] or k=v wins, and unifying all six sites on one format, is also left out — that's a deliberate output change.

Checklist

• Docs are autogenerated from CLI help — no help text changes in this PR
• Helm chart (charts/k8s-reporter/) — not needed
• Terraform provider — not needed

[claude]

Claude finished @dangrondahl's task in 52s —— View job

---

Follow-up review of 1a1dbdea..3ffbec84

This commit only trims code comments in tableHelpers.go and tableHelpers_test.go (dropping historical narration, keeping the durable "why") — no production logic or test assertions changed.

Status of previously raised findings:

1. getControl.go — duplicate unbracketed-tag formatting → resolved (unchanged by this commit; thread already resolved).
2. tableHelpers.go — guard comment contradicted the PR description → still resolved. The comment is further trimmed but remains accurate (no longer claims the API omits the key, no longer over-explains — states the guard's effect only). Thread already marked resolved.
3. getRepo.go — wrapper duplicated by two callers → resolved (unchanged by this commit; thread already resolved).
4. tableHelpers_test.go — test name/assertions → still resolved. Test name and require.Equal assertions are unchanged; only the doc comments above TestTagRenderingIsSortedAcrossPrinters and TestPrintEnvironmentAsTableWithoutTags were shortened. Thread already marked resolved.

No new bugs, regressions, or inaccuracies found in the lines this commit touched — the trimmed comments still correctly describe the code's behavior. 0 new inline comments.

[claude]

Follow-up review of d4d02be7..1a1dbdea

All four previous findings are resolved by this commit, and I've marked those review threads resolved:

1. getControl.go — duplicate unbracketed-tag formatting → resolved. getControl.go now calls the new formatPlainTags helper instead of re-joining sortedTagPairs itself.
2. tableHelpers.go — guard comment contradicted the PR description → resolved. The comment now says responses carry an empty tags object today and that the guard is defensive (matching the PR description), instead of claiming the key is omitted.
3. getRepo.go — wrapper duplicated by two callers, comment referencing sibling code → resolved. formatRepoTags is deleted entirely; getRepo.go and listRepos.go both call the new formatPlainTags in tableHelpers.go, which now has a self-contained doc comment (no cross-reference to formatTags).
4. tableHelpers_test.go — test name/assertions → resolved. Renamed to TestTagRenderingIsSortedAcrossPrinters, and all four subtests now assert the full rendered output with require.Equal instead of require.Contains on a fragment.

No new bugs or regressions found in the lines touched by this commit — sortedTagPairs is correctly reduced to two in-package callers (formatTags, formatPlainTags), and no dead code or unused imports remain from the formatRepoTags removal.

View job run