fix: discard without snapshots, move SLI to settings, fix version table

[TerrifiedBug]

Summary

• Discard changes: Now works for pipeline versions created before snapshot support was added — restores globalConfig, skips node/edge restore gracefully instead of throwing an error
• SLI badge: Removed from toolbar (was cluttering it). Per-SLI health status indicators (green/red/gray dot) now show in Settings > Health SLIs
• Version history: Fixed column header overlap ("Changelog" and "Created" were touching) caused by max-w-0 on the Changelog column

Test plan

• Deploy a pipeline, make config changes, verify discard works
• Open Settings gear > Health SLIs, verify per-SLI health dots show
• Open version history dialog, verify column headers are properly spaced

[greptile-apps]

Greptile Summary

This PR addresses three independent bug fixes in the pipeline editor: graceful handling of pre-snapshot discardChanges, relocation of the SLI health indicator from the toolbar into Settings, and a CSS fix for the version history table column overlap.

• Discard without snapshots (pipeline.ts): The old code hard-blocked discards if nodesSnapshot/edgesSnapshot were absent. The new code degrades gracefully — globalConfig is always restored; node/edge restoration is skipped when no snapshot is present. The transaction structure and existing withTeamAccess + withAudit middleware are preserved correctly.
• SLI health indicator move (pipeline-settings.tsx / flow-toolbar.tsx): The aggregate SLI badge is removed from the toolbar; per-SLI green/red/gray status dots are now shown inline in Settings > Health SLIs. A behavioral regression was introduced: the health query enabled condition is weaker than the original (slis.length > 0 vs. slis.length > 0 && !isDraft && !!deployedAt), which causes unnecessary failed requests on undeployed pipelines that have SLIs defined.
• Version history column fix (version-history-dialog.tsx): max-w-0 removed from the Changelog column header and cell; Version column widened from 80px to 120px. The truncation is now applied via max-w-[200px] on the inner span, which is the correct approach.

Confidence Score: 4/5

• Safe to merge — all three fixes are correct and well-scoped with one behavioral regression in the health query guard that should be addressed before merge.
• The backend discard logic is sound, authorization middleware is intact, and the UI fixes are straightforward. The only finding is that the relocated health query in SliSettings omits the isDraft/deployedAt guard that the original toolbar version had, which causes unnecessary failed API calls for undeployed pipelines but does not cause crashes or data loss.
• src/components/flow/pipeline-settings.tsx — health query enabled condition is weaker than the original.

_{Last reviewed commit: e1fd3c2}

[greptile-apps]

Health query fires on draft/undeployed pipelines

The original toolbar version of this query guarded with !isDraft && !!deployedAt before enabling the health endpoint. The new SliSettings component only checks slis.length > 0, so if a pipeline has SLIs configured but has never been deployed (or is still a draft), this query will fire and likely return an error or empty payload from the server.

The UI handles the missing data gracefully (healthQuery.data?.slis ?? []), so there's no crash — but it will generate unnecessary failed requests every 30 seconds as long as the Settings panel is open.

Consider passing isDraft and deployedAt into SliSettings and mirroring the original guard:

Suggested change

	const healthQuery = useQuery(
	trpc.pipeline.health.queryOptions(
	{ pipelineId },
	{ enabled: slis.length > 0, refetchInterval: 30_000 },
	),
	const healthQuery = useQuery(
	trpc.pipeline.health.queryOptions(
	{ pipelineId },
	{ enabled: slis.length > 0 && !isDraft && !!deployedAt, refetchInterval: 30_000 },
	),
	);

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/components/flow/pipeline-settings.tsx
Line: 334-338

Comment:
Health query fires on draft/undeployed pipelines

The original toolbar version of this query guarded with `!isDraft && !!deployedAt` before enabling the health endpoint. The new `SliSettings` component only checks `slis.length > 0`, so if a pipeline has SLIs configured but has never been deployed (or is still a draft), this query will fire and likely return an error or empty payload from the server.

The UI handles the missing data gracefully (`healthQuery.data?.slis ?? []`), so there's no crash — but it will generate unnecessary failed requests every 30 seconds as long as the Settings panel is open.

Consider passing `isDraft` and `deployedAt` into `SliSettings` and mirroring the original guard:

```suggestion
  const healthQuery = useQuery(
    trpc.pipeline.health.queryOptions(
      { pipelineId },
      { enabled: slis.length > 0 && !isDraft && !!deployedAt, refetchInterval: 30_000 },
    ),
  );
```

How can I resolve this? If you propose a fix, please make it concise.