feat(emf): model the Resolved Deployment target metamodel in Ecore

[ExtraToast]

Closes #87. Part of #78.

What this branch does

Task 2 transforms into the Resolved Deployment metamodel and Task 3 generates from it, so it is due for Task 1 alongside the source metamodel it is the other half of. Nothing existed: bundles/resolve and bundles/render each held a placeholder notes.ecore, and the module table named a target .ecore that was not in the tree.

• The metamodel, bundles/metamodel/model/resolved-deployment.ecore, 35 classifiers in one package. Its Java is generated into target/ by a second MWE2 workflow, the way the source metamodel's already is, and is left out of formatting, coverage and mutation on the same rule. plugin.xml registers the package, so an examiner's Eclipse import resolves it.
• models/minimal.resolveddeployment, the hand-written target model. Every value of it is derivable from notes.project.yml and the pinned Platform document and node contract beside it. It is test input and not an oracle: it is what the Acceleo templates are first run against (Tracer bullet for Task 3: render the minimal case with Acceleo from its hand-written target model #94) and what the QVT-Operational transformation must produce (Tracer bullet for Task 2: resolve the minimal case with QVT-Operational #90), so it is the target half of both tracers before either exists.
• DependencyEdges, the export of the parity contract's dependencies.json: the consumer Process, the provider Application and Surface, the address the consumer is given, and the peers the policy allows.

The decisions

The package holds two halves, and that is the point. The first is every decision layer 2 takes, in the words CONTEXT.md gives them. The second is one class per resource family of the project proposal's generated-resources table, each carrying the path it is written to. The production implementation keeps those apart because nothing there reads them together; a model-to-text template reads one model, so here they are one package. That difference is exactly why the parity contract compares the two implementations through dependency edges and generated files rather than through resolved.json.

No name crosses between the halves. A WorkloadFile carries the controller kind and the strategy that spell a Process's cutover, and the cutover itself stays on the Process. The template decides nothing, which is the layer rule holding one level further down than the chapter states it.

The target package stays out of the descriptor, and that is now checked rather than asserted. A target class reaching the descriptor would put this implementation's shape into an oracle the other one has to match. The check allows exactly three names through: Cutover, Match and DurabilityClass are the source metamodel's own vocabularies, carried over because they mean the same thing.

What the tests hold

Eight, and two of them tie the model to the tree rather than to itself: the eight paths it assigns are compared with the files actually on disk under spec/v1/examples/minimal/rendered/, so a template that writes somewhere else fails here rather than at review.

minimal holds no dependency edge, which is the point of that case, so the edge's own shape is proved on a model built in the test: one edge the policy names a peer for, and one that resolves to an address with no peer of its own.

The pending witnesses move

REQ-033's reason said there was no Ecore metamodel here to validate against. This change makes that false, so it would have become a lie on merge. What is still missing is a model this implementation derives, which is #90. REQ-034 waits on placement resolving against the node contract (#91), and REQ-035 on something rendering (#94). All three stay owed; none is quietly discharged.

Verification

./mvnw -B -ntp verify from emf/ exits 0 locally: 53 tests, the jacoco floor at 1.00 and pitest's mutation threshold both met. Root npm run verify green at 782 tests, 35 ledger rows, 99.2% statements.

Two failures were caught locally before anything was pushed, which is worth recording because the previous branch learned it the hard way: spotless formatting, and the jacoco floor, because the edge-building path never ran against a case that has no edges.

[github-actions]

Shape

bucket	files	additions	deletions
production code	4	+453	-0
tests	1	+226	-0
documentation	2	+18	-3
tooling	4	+156	-7

Coverage

Baseline: main @ 1061212.

metric	this branch	main	diff
statements	99.20%	99.20%	0.00
branches	96.56%	96.56%	0.00
functions	100.00%	100.00%	0.00
lines	99.13%	99.13%	0.00

This pull request changed no covered line.