fix(code-block)!: errors following a full review of the component (#DS-5482) - #1988

`maxHeight` was published as `InputSignal<number>` over an `undefined!` default,
so a code block with no `[maxHeight]` binding reported `undefined` from a
non-nullable type — an assignment to a `number` held `undefined`, and any
arithmetic on it produced NaN. It reports `number | undefined` now.

`KbqCodeBlockHighlight.file` was a write-only required input: a setter with no
getter that kicked off highlighting as a side effect. It is a required signal
input driven by an effect now, so it can finally be read.

The `max-height` applied while `viewAll` is off is a `computed`. It was a getter
read from a `[style.max-height.px]` binding, so it only re-evaluated when
something else marked the view dirty.

`softWrap`, `viewAll`, `canDownload`, `activeFileIndex` and `files` are backed by
signals. They stay accessor inputs with the same types and the same two-way
outputs: the component writes them as well as the binding, and a `model()` cannot
carry the `booleanAttribute` / `numberAttribute` transform they need.

BREAKING CHANGE: `KbqCodeBlock.maxHeight` reports `number | undefined` instead of
`number`, and `KbqCodeBlockHighlight.file` is a readable required signal input
rather than a write-only setter. Reported by the
`code-block-optional-max-height` schematic.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

`files.length < activeFileIndex` left the first out-of-range index in place:
three files with an active index of 2, shrunk to two, kept index 2 and the
template rendered `files[2].content` from undefined. The new test fails against
the old comparison and passes against `<=`.

The review's own narrowing fix hoisted the `maxHeight` read out of
`checkOverflow`, freezing the threshold at `ngAfterViewInit` so the resize
observer compared against a stale value. The read is back inside the callback.

`KbqCodeBlockHighlight` started a second highlight.js import for any `file` that
arrived while the first was still loading, registering the line-numbers plugin
twice, and neither continuation was cancelled on teardown. Deduped behind
`loadOnce()` with an `onCleanup` guard, and the promise is cleared on failure so
a later file change can still retry.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

`hideTabs` is derived instead of written. A single file with no filename still hides the tab bar -
the rule the input has always documented - but the component no longer writes `true` into its own
input to get there. The write latched the bar off for the life of the component, so naming the files
later never brought it back and every file past the first stayed unreachable, and it re-emitted
`hideTabsChange` on every `files` assignment even when nothing had changed.

`[hideTabs]="false"` no longer shows the bar for a lone unnamed file. That case only ever worked
because the consumer's template happened to list `[files]` before `[hideTabs]`: with the attributes
the other way round the same code hid the tabs, so what the old test pinned was attribute order
rather than a contract.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The reseeded migration lost two reports the schematic it replaced emitted: `maxHeight` reads and
`KbqCodeBlockHighlight.file` writes matched neither member set, so the only mention left was a line
in the once-per-project summary with no file attached - while the guide still tells consumers to fix
exactly those call sites. Both are reported per file again, and the receiver pass resolves the
highlight directive as a type of its own.

`WRITABLE_MEMBERS` aliased `SIGNAL_MEMBERS`, so the flag threaded through `classifyAccess` was always
true; the set and the parameter are gone.

`onSelectedTabChange` compared the clicked index against the rendered one, so clicking the tab that
is already on screen left an out-of-range `activeFileIndex` in place - and a later `files` that grew
past it would jump there with no user action. It compares against the bound index instead.

The overflow gate read `pending()` in the same effect that owns the ResizeObserver subscription, so
every highlight tore the subscription down and rebuilt it around two forced layout reads. The
subscription follows the element, a second effect retakes the measurement when highlighting settles.

Tests for the paths the refactor changed and nothing covered: a valueless boolean attribute, the
deprecated `canLoad` turning the download button on, and `codeFiles` filling in while `files` is
empty - the shape `docs-live-example-viewer` binds.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The backing inputs were `protected`, which compiles inside the library and breaks every consumer:
with `strictTemplates` a `[softWrap]` binding is checked against the class member the alias points at,
so `docs-examples` failed with `TS2445: Property 'softWrapInput' is protected`. `ng build components`
never catches it - the component's own template does not bind its own inputs - and `ts-jest` does not
type-check templates at all, so the local run was green. `ng build docs-examples` is the check that
reproduces it, and `tree/toggle.ts` keeps its aliased `disabledInput` public for the same reason.

`PROTECTED_MEMBERS` in the migration is renamed `PLUMBING_MEMBERS`: the members it lists are public
now, and its report says what they actually are - read-only inputs that carry the attribute and its
transform, and report what was bound rather than what the component holds.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

The "What it does not do" heading had the paragraph glued onto it and the rest orphaned after a blank
line. The section also lost the sentence about what is reported rather than rewritten: the edit that
was supposed to add it used a plain `String.replace` whose pattern no longer matched after an earlier
edit, so it silently did nothing. Heading and paragraph restored together.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>