[Feat] Add applicability for prompt-dependent artifact/message scenarios

## Environment

- TCK repository: `a2aproject/a2a-tck`
- Tested commit: `5996b79f9cefa6fc390980e383e358a66fb9e49e`
- Level: `must`
- Transports: JSON-RPC and HTTP+JSON
- SUT: an A2A 1.0 narrow travel-search agent that requires exact dates, occupancy and a hard EUR
budget before it may create a priced result

## Reproduction

Run the unchanged official TCK against a conformant narrow agent that returns a typed
`input-required` task for the literal prompt `TCK artifact test`:

```text
uv run python run_tck.py --sut-host --transport jsonrpc --level must -v
uv run python run_tck.py --sut-host --transport http_json --level must -v
```

The same five compatibility tests fail on both transports:

```text
tests.compatibility.core_operations.test_artifacts.TestTextArtifact::test_task_has_text_artifact
tests.compatibility.core_operations.test_artifacts.TestFileArtifact::test_task_has_file_artifact
tests.compatibility.core_operations.test_artifacts.TestFileUrlArtifact::test_task_has_file_url_artifact
tests.compatibility.core_operations.test_artifacts.TestDataArtifact::test_task_has_data_artifact
tests.compatibility.core_operations.test_artifacts.TestMessageResponse::test_returns_message_with_text_part
```

All five tests send the same generic text, `TCK artifact test`. Their generated `messageId` values
use these scenario prefixes: `artifact-text`, `artifact-file`, `artifact-file-url`,
`artifact-data`, and `message-response`.

## Actual behavior

The SUT returns a valid A2A task in `input-required` state with typed missing-field details and a
text fallback. It does not invent travel inventory, dates, a budget, a file, a URL or an unrelated
artifact. Transport, Agent Card, content-type and normal task lifecycle checks pass.

Redacted response fragment, equivalent on both transports:

```json
{
"status": {
"state": "TASK_STATE_INPUT_REQUIRED",
"message": {
"parts": [
{
"data": {
"error": "MISSING_REQUIRED_FIELD",
"message": "exact input is required before searching",
"details": {
"missing": [
"checkin and checkout as exact ISO dates",
"hard EUR budget",
"one- or two-night duration"
]
}
}
}
]
}
},
"artifacts": []
}
```

At the pinned revision, the public test totals are:

- JSON-RPC: 58 passed, 5 failed, 172 skipped;
- HTTP+JSON: 54 passed, 6 failed, 175 skipped.

The additional HTTP+JSON failure is the separately reported streamed-error-body defect in
[issue #225](https://github.com/a2aproject/a2a-tck/issues/225), with a proposed fix in
[PR #226](https://github.com/a2aproject/a2a-tck/pull/226).

## Expected behavior

Protocol conformance should not require a domain-specific agent to fabricate the artifact or direct
message implied by a generic magic prompt. One of these applicability mechanisms would keep the
tests useful without weakening protocol checks:

1. allow the SUT manifest to declare unsupported scenario fixtures;
2. treat a valid `input-required` task as not applicable for these product-semantic scenarios;
3. make artifact/message fixtures opt-in capabilities rather than unconditional MUST expectations;
4. split protocol-shape validation from the prompt-dependent example SUT assertions.

The report does not request a waiver for malformed messages, invalid task states or broken
transports. It requests only a machine-readable applicability path for prompt-dependent product
semantics.

## Evidence hygiene

The minimal reproduction above contains no credential, provider payload, affiliate identifier,
private URL, personal data or customer prompt. Raw reports remain in the private CI artifact store.