chore(metamodel): regenerate to fix the dropped typeof:/abstract: keys

[kervel]

Lands the fix for #108 by regenerating the metamodel, and corrects the reproducibility rule
#112 introduced.

The criterion was wrong

GENERATED_FROM listed the generator commits not in linkml/main and declared that the crate
"cannot be reproduced from upstream linkml alone" until they merged. That made every regen wait
on an upstream review queue for no reason.

Naming the changes is enough. Each extra commit is now resolved to every PR whose head carries
it, and the stamp reads as a recipe:

generator_revision = e7015ec5cfe365db14e4aea92b646f58759ef254
generator_describe = v1.11.1-288-ge7015ec5c

# To reproduce: apply these generator changes on top of linkml/main,
# then run ./regen.sh. They need not be merged; naming them is enough.
# Newest first:
# e7015ec5c fix(rustgen): preserve LinkML slot name on the wire for escaped fields  (linkml/linkml#3919)
# 0e8b0cb3a fix(rustgen): include concrete base class in its *OrSubtype enum  (linkml/linkml#3919)
# a9c00ca0a fix(rustgen): emit serde tag for type designators on subtype enums  (linkml/linkml#3919)

So a generator fix has to be identifiable, not merged: pushed, with a PR open, so the stamp
names something a reader can fetch rather than a sha that only exists on the author's disk.
All three commits resolve to #3919 because they are stacked on that branch — the older PRs
#3661/#3662 carry pre-rebase versions with different shas, and fetching #3919 gets you all three.

regen.sh also warns when the generator checkout has uncommitted changes: the recorded revision
then does not determine the output, which is the one thing the stamp promises. A warning, not a
refusal — the diff may be unrelated to the generator.

The regen

typeof and abstract now line up with their typeof_/abstract_ fields instead of being
dropped, so a derived type has a parent again. The seven tests #112 left #[ignore]d go green
unchanged, which is what they were for, and the ignores are removed. Fixes #108.

User-visible, in turtle output:

-geotest:mirror "https://mirror.brussels.be" ;
+geotest:mirror <https://mirror.brussels.be> ;
-geotest:track_length "1200" ;
+geotest:track_length 1200 ;

Nothing is hand-patched. The renames are the generator's own output, derived generically from
"was this name escaped" rather than from a list of names — on meta.yaml that comes to exactly
five (typeof on TypeDefinition, abstract on four structs; type_ does not occur).

What rides along

Two generator commits sit under the fix on that branch, so regenerating brings them:

• The concrete base class becomes a variant of its own *OrSubtype enum —
  ExtensionOrSubtype::Extension, EnumExpressionOrSubtype::EnumExpression, with the matching
  From, PyO3, from_pair_*, stub and poly.rs dispatch arms (Track linkml/linkml#3662: *OrSubtype enums omit the concrete base class as a variant #110). This is the only reason
  the Python stubs move: every extensions field widens from Annotation to
  Annotation | Extension, 82 lines of exactly that substitution. stub_gen --check fails
  without regenerating them, so they are regenerated here.
• The type-designator serde tag (Track linkml/linkml#3661: type-designator subtype enums emitted untagged #111), which changes nothing in this crate — meta.yaml
  declares no type designator.

Verification

cargo test --workspace 74 suites green, cargo fmt --all --check, clippy with
-D warnings, and stub_gen --check all clean. ./regen.sh --check reports
OK: committed crate reproduces from v1.11.1-288-ge7015ec5c, so the gate works in both
directions — it reported drift before this regen and is clean after.

Note for whoever bumps a downstream pin

A consumer asserting the buggy behaviour on purpose will now fail, which is the intended signal.
In asset360 that is sparql_terms.rs::typeof_only_type_has_no_datatype_and_is_not_numeric:
trackLength becomes xsd:integer and numeric, so the assertion inverts.