build(release): publish the showcase the cut produced, not the one lying in target (#477)

* build(release): publish the showcase the cut produced, not the one lying in target

cut-release.ps1 never ran GenerateAllExamples. Run-ShowcaseSync installed the
train siblings, recompiled the examples so ShowcaseMetadata.GH_BASE pointed at
the tag, and then synced whatever happened to be sitting in
examples/target/generated-pdfs from earlier commands — on a clean checkout,
nothing, and the cut aborts mid-flight. Step 4 now regenerates the catalogue
first, with clean, so the published site is a function of the tag rather than of
the maintainer's shell history. It runs after the pom bump and the GH_BASE
rewrite, so version-stamped renders carry the new version.

ShowcaseSync copied but never removed, so an example that was renamed or deleted
kept its artifact live on GitHub Pages indefinitely — reachable by URL, absent
from the manifest, rendered from source that no longer exists. It now clears the
three output trees first. Three documents went with it: invoice-custom-theme,
whose example was removed, and invoice.pdf / proposal.pdf, which no example has
written since June. Only files are deleted, not directories: git tracks no empty
directory, the copy recreates what it needs, and removing them is what makes
this fail on Windows with AccessDeniedException.

Eight generated artifacts had no ShowcaseMetadata entry, so lookup fell back to
a filename-derived title, a generic description and a link to the examples root
— a placeholder card. Two of them were the flagships 2.1.x exists to show,
social-card and linkedin-carousel. The manifest now carries 90 cards and no
placeholders. EngineDeckV2Example, which renders the deck the README banner is
cut from, was never in GenerateAllExamples despite being linked from the README
and covered by a snapshot test; registering it made it the eighth.

invoice() and proposal() hardcoded their producing example, so the two v2
entries could not name theirs. Both now take the class like cv(), letter(),
feature() and flagship() already do.

Five committed previews were materially stale. pdf-chrome.pdf is the visible
one: the whole body is drawn at low alpha with the DRAFT watermark bleeding
through every panel, where a fresh render is solid. Staleness was decided by
rendering both copies and comparing pixels — the five differ by 11.9 to 36.4
mean channel delta, the other twenty-one by less than 0.16, which is the date
stamp in the corner and nothing else.

Eleven committed binaries totalling 9.01 MiB are referenced by no file in the
repository — assets/Overview.png alone is 5.8 MiB. The .gitignore allowlist
still whitelisted docs/assets, docs/showcase and site/, three trees with no
tracked file between them.

* docs(changelog): record the showcase and asset regeneration in v2.1.1

* build(release): give the Maven banner a PDF, and make the smoke test assert

The banner deck never reached the site. GenerateAllExamples called only
MavenBannerPptxExample.generate(), which writes flagships/maven-banner.pptx,
while the class also exposes generatePdf(). ShowcaseSync walks .pdf files, keys
one card per PDF, and copies a .pptx only as that PDF's same-named companion —
so a deck with no PDF twin lands in target and stops there, and the metadata
entry written for it goes unused. The banner appeared in web/examples.json zero
times. Generating the PDF first puts it on the site: the sync now reports 91
PDFs and 8 PPTX twins where it reported 90 and 7.

everyGeneratedDeckHasAPdfTwin states that invariant so the next deck cannot
repeat it. The smoke test around it asserted nothing — its own comment said "at
least one PDF should have been written" above assertThat(true).isTrue(). It now
reads the output tree: PDFs exist, every deck has its twin, and the catalogue
spans templates, features and flagships rather than one corner of it.

ShowcaseSync collected the PDF list after clearing the published trees, so a
generation run that succeeded while writing nothing would have deleted the whole
site and written an empty manifest, with the failure visible only once the site
was live. The list is now collected first and an empty one aborts before
anything is removed.

Author: DemchaAV

.gitignore

@@ -45,14 +45,6 @@ render-pdf/logs/
 *.pdf
 # Allow PDF previews that are committed README assets.
 !assets/readme/**/*.pdf
-# Allow PDF previews that ship with the GitHub Pages showcase site.
-!docs/assets/**/*.pdf
-!docs/showcase/**/*.pdf
-# Same whitelist for the Next.js showcase site under site/ — 53 PDFs
-# mirrored from docs/showcase by Sync-SiteShowcase, plus 3 Playground
-# preview PDFs (hello/invoice/cv) bundled with site/public/previews/.
-!site/public/showcase/**/*.pdf
-!site/public/previews/*.pdf
 # Same whitelist for the live GitHub Pages showcase under web/. The showcase
 # moved out of docs/ into web/ but this list was not updated, so every newly
 # added example PDF was silently ignored and 404'd on the deployed site.

CHANGELOG.md

@@ -22,6 +22,23 @@ follow semantic versioning; release dates are ISO 8601.
 - **`graph-compose-render-pptx` declares PDFBox.** It compiles against
   `org.apache.pdfbox` types while declaring only `fontbox`, taking the rest
   transitively; the resolved version is unchanged.
+- **The release publishes the showcase it just built.** `cut-release.ps1` never
+  ran `GenerateAllExamples`, so the site was synced from whatever happened to be
+  in `examples/target/generated-pdfs` — nothing at all on a clean checkout, which
+  aborted the cut. Step 4 regenerates the catalogue first.
+- **A removed example stops being published.** `ShowcaseSync` copied but never
+  deleted, so an artifact whose example was renamed or removed stayed live on the
+  site indefinitely, reachable by URL and absent from the manifest. Three such
+  documents are gone.
+- **Every generated document has a showcase card.** Eight artifacts had no
+  metadata entry and published as filename-derived placeholders linking the
+  examples root — among them both flagships the 2.1 line exists to show.
+  `EngineDeckV2Example`, which renders the deck the README banner is cut from,
+  was never wired into the example runner at all.
+- **Five committed previews were re-rendered.** The largest, the PDF-chrome
+  preview, was drawn at low alpha throughout with the watermark bleeding through
+  every panel. Eleven committed binaries that nothing references, 9 MiB of them,
+  are removed.
 - **The release script owns the README release-status block.** It was a
   hand-edit the script only validated, and the validation demanded the *target*
   version — so between releases `develop` had to advertise an unpublished version