Editor: imported pages bring their structure elements (#183)

[timzifer]

Fixes #183. Stacked on #225 (merge that first; this PR then shows only its own commit).

API

type ImportOptions struct {
	// …
	Structure bool // the structure elements whose content is on imported pages
}

What is brought (internal/pdfedit/structure.go)

• The elements of the source's /StructTreeRoot with content on an imported page (marked content, MCRs, OBJRs), with the elements above them; an element left without content is dropped.
• Placement. Where the result has a tree, the elements follow its own in a /Part under its single /Document element, or under its root when it has none: the source's own /Document becomes the Part, or a Part is made around its top elements. The Part gets the source catalogue's /Lang where it differs from the result's. A result without a tree (NewEditor, or an untagged edited document) takes the first tree brought as its own; further sources become Parts under its Document.
• Keys. /StructParents of pages, /StructParent of their annotations and /StructParents//StructParent of the form XObjects they draw are renumbered from the result's /ParentTreeNextKey; the parent tree is written anew as one flat number tree (the result's entries plus the new ones). A page imported twice is tagged once, its first copy (the second copy and its annotations have no keys). A form XObject that only pages which did not come draw as tagged content loses its keys rather than keep a dangling one.
• References. /Pg, MCR /Stm, OBJR /Obj (the first copy of the annotation) and /Ref are mapped to the copies; references to elements that are not kept become null or are left out.
• Maps. /RoleMap merged; where the result maps a type otherwise, the copied elements are written with what the source means by it (Heading → H1 under a result that maps Heading → P). /ClassMap merged, a class defined otherwise renamed Cls_2 (in /C too). /IDTree rebuilt with the kept elements, an identifier the result has renamed id_2. /Namespaces joined.
• /MarkInfo gets /Marked true, the result's other entries kept.
• Pages of the edited document imported into itself bring nothing more, as for outline and destinations.

Deviation from the issue's text: the parent tree is written as one flat /Nums array instead of being extended in place; number trees cannot take more entries without their /Limits being rebuilt, as for the destinations tree in #135.

Tests

• TestImportStructure: into a tagged document (role, class and ID conflicts, Lang of the part), into an untagged one, into a new one, from two documents into a new one; save and update. Every case checks the tree as Document.Structure reads it and the parent tree against the elements (parentTreeErrors: every page, annotation and form key leads to the elements that name its content, every entry is used, keys below /ParentTreeNextKey).
• TestImportStructureTwice, TestImportStructureOptimised (object streams, linearised, pruned, joined, compressed), TestImportNoStructure (unchanged default).
• Corpus API run (cmd/corpus run -api) gets the step import structure: the pages looked at, imported with their structure into the document itself (opened a second time), must keep as much marked content in the tree as the originals. pdf.js corpus, 3 pages per file: no difference; only the refusals every other save step has (signed, no assembly permission).
• go test ./..., -tags cera_nogyre green locally; lint clean but the CRLF noise.

Docs: docs/features.md updated (editor list, "not done" list, tagged PDF row).