The live IT compiles against a client generated from either contract

[gsavastano]

What

Makes OpenDppLiveIT compile against a generated client from either contract, by not naming the two
things that differ by contract at compile time.

• resolve(api, name, key) finds a resolver by NAME and passes null for every parameter after the key.
  Those are all optional on the wire — a grant token, and since contract 1.16.0 the representation flag
  — so the request is exactly the one the two-argument form made: anonymous tier, default compressed
  document. InvocationTargetException is unwrapped, so the 404 case still asserts on a typed
  ApiException rather than a reflection wrapper.
• The getMetadata() assertion is split. What it actually tested — untyped JSON surviving Jackson — is
  now asserted through @context, which every contract declares as a relaxed Object, by requiring a
  structured value rather than a flattened string. The non-empty map check is kept opportunistically via
  optionalMap, so contracts up to 1.15.0 still prove the body is non-empty; from 1.16.0 the property is
  gone and the body's elements are root members with no generated getter, so there is nothing to read.

Test-only. No regenerated client, no version bump, no vendored spec change — those stay the mirror sync's
to write.

Why

This file is hand-written against the generated client, and two lanes generate that client from two
different specs: this repo's CI from the vendored openapi.json, and opendpp-node's SDK regen gate from
its live contract. They can sit a contract apart, so any hand-written reference to a generated signature
or getter compiles in one lane and breaks the other.

Contract 1.16.0 did exactly that. It added representation to both public resolvers and moved the
passport body out of a metadata object onto the document root. This file kept compiling here against
the vendored 1.15.0 and stopped compiling in node's gate: four call sites on argument count, one on the
missing getter. That gate feeds the required rollup on node's release/train-*, so a contract change was
blocked by a test in another repository that could not have been green in both places at once.

The header now states the constraint, so the next contract change does not rediscover it: anything
asserted unconditionally has to be true of both contracts.

Verification

Both directions, using node's own gate script (scripts/sdk-regen-verify.ts <checkout> --only=java),
which is what its CI runs:

• Against a client generated from the live 1.16.0 contract: failed on exactly the five reported
  errors before this change; after it, regenerates, builds and tests clean at 1.16.0.
• Against the vendored 1.15.0 in this tree: ./gradlew build passes, unchanged.

The live IT itself stays opt-in behind OPENDPP_LIVE_TEST=1, so neither run executed it against the
hosted node.

Note for the merge

node's SDK regen job checks this repository out at ref: main, so the gate only clears once this is on
main.