Bring the free canvas fragments onto one shape

[ctgnz]

Closes #78.

A free canvas icon holds a <g>#main, a <g>#template, zero or more <g>#example groups, and whatever <defs> the main group needs to render. 545 of 577 already were that. This brings the stragglers into line so a consumer asking "which element is the content" has one answer rather than four.

Scoped down from the issue's original three options after discussion: the scaffolding stays in the fragment for now, because FREE_CANVAS icons will need separate handling anyway.

Two commits

1 — Normalise the free canvas fragments edited in Inkscape. The content changes in these four are CTG's: each Cyberspace fragment now keeps only the drawn control measure in #main, with the illustrative endpoints and net boxes moved into a hidden example group and the anchor points into the template. Inkscape rewrites a file wholesale on save, so all four needed the normaliser afterwards — 0 refused, which is the guarantee wanted: the drawn content is byte-for-byte what CTG left, and only the editor's leavings came out.

One follow-on belongs with that rather than the mechanical pass: 60200100's example group carried id="g15", which the normaliser correctly reads as a generated id and drops, leaving it anonymous. It's now id="example".

2 — Bring the free canvas fragments onto one shape. Nine fragments:

problem	files
id on a bare drawing element, no group	25218800, 25281901 (polygon#main), 25200600, 25200700 (loose rect)
scaffolding group named template2 when it is an example	25200600
<pattern> loose at the root	25240301/2/3, 25172000, 25270400
<pattern> inside #main itself	25240301, 25270400

Patterns move into a <defs>. Ordering and ids otherwise left exactly as found, with one exception in a third commit: 25200600's hidden #474747 rectangle is not part of the construction guide but a backdrop for the white border of the symbol in #main, so its group is renamed template2 → example. The corpus agrees — three fragments carry that backdrop and the other two, 25200400 and 25200700, both file it under an example group.

Nothing changes what is drawn

A <g> carrying only an id paints exactly as its child did, and a <pattern> renders only where something references it — and nothing does. The whole tree holds one live url(#…) reference, 25132000's marker, untouched. The issue's note that the pattern files hold the only internal reference is no longer accurate.

SvgFingerprint can't guard this: it records structure, so inserting a group or moving an element changes it by design. Verified two independent ways instead:

• Every drawing primitive compared before and after with its ancestors' attributes folded in, so an added group is invisible to the comparison, and each pattern subtree compared on its own. All nine identical.
• FragmentMeasurer re-run over all 577 free canvas icons — every bound unchanged. The only bounds that moved in this PR are 60200100 and 60200200, from CTG's edits.

mvn verify green, Spotless clean.

Left alone deliberately

• 26 fragments have no template group at all (22 are #main alone). That's a question about what those files draw, not about their shape.
• 25270400's two patterns have no id, so they can't be referenced from defs either. Naming them belongs with the pattern usage work.
• Group order varies — 338 are main + template + example, 13 put example before template. Reordering changes paint order, so it stays out of a "no visual change" pass.

After this, 551 of 577 hold only defs/main/template/example*, and the remaining 26 differ only by a missing template.

The matching changes to jmsfx-hallux are in the hallux working tree, uncommitted.

🤖 Generated with Claude Code