chore: neutral simplification markers and #5707 review follow-ups

[nabramovitz]

Three tidy-ups off the back of the #5707 review. No behaviour change; the only production-code edit is a type narrowing that removes a redundant operator.

1. Neutral markers for deliberate simplifications

Six comments flagged deliberate shortcuts or do-not-refactor notes behind an ad-hoc prefix instead of the marker already used elsewhere in the tree. This renames the prefix; the notes themselves are unchanged in substance and worth keeping — one of them is load-bearing.

File	New marker	Why
plan-parameters-preview.util.ts:19	simplification:	strips only top-level $schema, names the upgrade path
schema-resolve.util.ts:168	simplification:	empty strings treated as unset, deliberately scoped out
tailwind-dialog.service.ts:412	simplification:	hand-rolled drag instead of cdkDrag, with the reason
tools/stb/src/main.ts:132	simplification:	naming-convention detection, names the upgrade path
schema-resolve.util.ts:99	NOTE:	recursion guard, not a shortcut
tailwind-snackbar.service.ts:207	NOTE:	inverse-overlay warning, not a shortcut

simplification: is not a new convention — scripts/lint-e2e-data-test.mjs:68 already uses it. The two that became NOTE: were never simplifications: one is a correctness guard against infinite recursion on self-referential $ref schemas, the other warns that the snackbar's bg-gray-800/text-white are a deliberate inverse overlay kept raw so a theming sweep doesn't convert them to content-* tokens and produce light-on-light text.

2. Follow-ups from the #5707 review

Three loose ends raised in review on #5707 that merged before they were picked up.

onNext coalesced the endpoint guid at the saveAppDetails call with ?? ''. SourceType.endpointGuid is optional, so that operator was carrying the type rather than ever changing a value — the public branch is already guarded non-empty above it and the other branch is a literal ''. Narrowed at the declaration instead and dropped from the object literal.

The two setGitMode specs asserted with toHaveBeenCalledWith, which matches any call, while their comment described the last one. Switched to toHaveBeenLastCalledWith so the assertion states what the comment claims — a strictly stronger check, still green.

The deferral comment in setActive said a synchronous pOnEnter would run "before the entering step is marked active". The step is marked active a few lines above it. The real constraint is that change detection has not yet re-rendered the content outlet for the new currentIndex, so the callback would run against a view still showing the outgoing step. Reworded to say that.

3. Real coverage for projected step content

Closes #5710.

The existing stepper suite builds StepComponent instances by hand and assigns allSteps directly. That covers setActive's bookkeeping but never renders a stepper, so nothing in it can observe how projected step content is actually instantiated — which is how the incorrect claim in section 2 went unchallenged in the first place.

Adds a host that mounts a real <app-steppers> with projected content and pins two behaviours:

• A step that is never activated still has its content constructed and change-detected on first render. Ivy does not defer <ng-content> inside an <ng-template>, so a step's directives and their ngOnChanges run up-front. The probe places its directive behind nested @if blocks to mirror deploy-application-step2, where precisely this caused the regression fixed in fix(deploy): make GitHub/GitLab app deploy work (Source Config, private repos, token re-validate) #5707.
• onEnter still arrives only after the activating render, asserted through a rendered stepper rather than hand-built objects.

Both were checked against mutations rather than assumed meaningful: disabling the probe's outer @if fails the first (while construction is still logged, so it distinguishes template execution from construction), and calling pOnEnter synchronously fails the second — identically to the pre-existing hand-built test.

The older hand-built spec is left as-is; it remains a reasonable unit-level check of setActive. The point is that it is no longer the only coverage.

---

make check gate green.

[norman-abramovitz]

LGTM