docs(examples): the catalogue stops calling current code legacy

[DemchaAV]

Why

Two entries in the example gallery sat under this heading:

🗄️ Legacy — pre-rebuild examples kept for downstream callers still on V1; do not start new code here. See docs/templates/which-template-system.md for the V1 → V2 migration path.

There is no V1. docs/templates/which-template-system.md says in its second sentence that GraphCompose 2.0 ships no other template system, both examples compile against 2.1.1-SNAPSHOT on the current DSL, and three hundred lines below the same file describes one of them as a "useful starting point when your domain doesn't fit any built-in template". A reader looking for exactly that — hand-composed, no preset — was told not to start there.

Separately, the run instructions led with GenerateAllExamples. That is not a cheap first command: it is a hand-written list of a hundred generate() calls preceded by ExampleOutputPaths.clean(), so a reader following the page in order emptied the output directory and rendered the whole catalogue before seeing one document.

What changed

• The category is now ✍️ Hand-composed — what those examples are — with the V1 wording and the migration link gone from both the legend and the section heading. The #handcrafted-proposal and #weekly-schedule sub-anchors are untouched, so nothing that links into the file breaks.
• Run instructions reordered: one example first (ModuleFirstFileExample, a Start-here row), then the batch entry point below it, labelled as the way the committed previews and the showcase site get refreshed.
• Release labels the docs(examples): stop the examples describing the releases they were written for #491 sweep left behind are gone: ## v1.5 feature showcases, "all v1.8 native vector charts", "the v1.9 book primitives", "byte-identical to v1.4", and "v1.5 cinematic features" in the intro. Also XxxTemplateV2, a wrapper name that never existed under that spelling and no longer exists under any.
• docs/recipes.md: the extension-guide entry no longer dates itself to v1.5, and a 1.x migration guide is dropped from the current cookbook's further-reading list — it remains reachable from the documentation index's Historical documentation section.

What I did not do

The recon for the previous PR noted that four Start-here rows (Engine Showcase, Engine Deck, Twin Output, Maven Central Banner) are bold labels rather than in-page links, unlike the four above them. That is not an oversight: grep '^### ' shows those four examples have no detail section in the file to link to. Giving them anchors means writing four new sections, which is a different change from this one.

Verification

./mvnw -B -ntp clean verify — BUILD SUCCESS, exit 0, 692 tests in the closing module. DocumentationSnippetCompileTest 6 (this file carries 39 java fences, all still marked), CanonicalSurfaceGuardTest 11 (every relative link resolves), DocumentationCoverageTest 5.

grep -n "v1\.[0-9]\|\bV1\b\|Legacy" over examples/README.md returns nothing.

One thing worth an eyeball rather than a test: the new pointer [🚀 Start here](#-start-here) is the only link in the repository to an emoji heading, and no test validates in-page anchors — CanonicalSurfaceGuardTest skips # targets by design. The slug follows github-slugger's rule (emoji stripped, leading space becomes the leading hyphen), but the rendered page is the only place to confirm it.

Lane: docs. No production code, no public API.

[DemchaAV]

Both fixed, and both deserved the catch — a PR about instruction accuracy that shipped two inaccurate instructions in the run path is the worst place for them.

1. The class-name rule is now stated and proven. The text gives the derivation explicitly — drop src/main/java/ and .java, swap / for . — with the worked example you suggested. Then I checked it generalises rather than assuming: every one of the ~40 source paths linked from this file declares a package matching its directory (zero mismatches), and each Start-here class has a main. Finally I ran it end to end on a row I had not pasted — derived com.demcha.examples.templates.invoice.InvoiceCinematicFileExample by the rule, ran it, exit 0, Generated: …invoice-cinematic.pdf, one document.

2. GenerateAllExamples publishes nothing — corrected. Verified in the source before rewriting: it writes only into examples/target/generated-pdfs/; ShowcaseSync copies into web/showcase/, renders the PNG thumbnails and rewrites web/examples.json; and Refresh-CommittedPreviews in cut-release.ps1 walks the generated tree and re-copies the already committed subset under assets/readme/examples/. So the batch runner is the source both of those read from, which is what the text says now. The same wrong sentence was in the CHANGELOG entry and is fixed there too.

I left the ShowcaseSync command out of the block: this section is the reader's path to running an example, and that tool is maintainer-side, driven by the cut. Naming it and what it does seemed the right amount. Say the word if you want the command shown too.

./mvnw -B -ntp clean verify — BUILD SUCCESS, 692 tests in the closing module.

Head is now 06befc73.