Validate ingestion support for indexed schemas

[jwils]

Why

Missing ingestion support should fail before events reach the indexer. Artifact generation currently accepts schemas with no ingestion extension, and the JSON-aware indexer accepts configurations without JSON artifacts or an adapter.

What

Require indexed schemas to register an indexer extension that defines ingestion_adapters_by_format. Reject construction of JSONIngestion::Indexer unless its base indexer has JSON schema artifacts and a json ingestion adapter.

How

Runtime metadata generation checks the registered extension methods without constructing runtime services. The indexer still checks for an empty adapter registry at runtime.

The JSON wrapper checks for versioned JSON schema artifacts and reports an actionable configuration error when JSON support is absent. Tests use a shared non-JSON adapter whose registration is verified through generated indexing operations.

Risk

Schemas that register only unrelated indexer extensions now fail artifact generation. The schema check verifies the extension interface; it does not instantiate adapters or validate their runtime configuration.

Testing

No manual testing. Local datastore tests were blocked because Docker Desktop requires sign-in to the squareup organization.

Bigger picture

This follows the ingestion adapter extraction in #1302 and the JSON-aware indexer in #1220. It addresses the review feedback deferred from #1220.

[myronmarston]

indexer.schema_artifacts.respond_to?(:available_json_schema_versions)

This reminds me--I had designed ElasticGraph::SchemaDefinition::Results to have the same interface as ElasticGraph::SchemaArtifacts::FromDisk so that the in-memory results object could stand in for a FromDisk instance of schema_artifacts in our tests so that tests can operate entirely in memory w/o disk in most cases.

But currently they diverge:

• FromDisk still has JSON schema APIs on it:

elasticgraph/elasticgraph-schema_artifacts/lib/elastic_graph/schema_artifacts/from_disk.rb

Lines 66 to 111 in 2932af6

	def json_schemas_for(version)
	unless available_json_schema_versions.include?(version)
	raise Errors::MissingSchemaArtifactError, "The requested json schema version (#{version}) is not available. " \
	"Available versions: #{available_json_schema_versions.sort.join(", ")}."
	end
	json_schemas_by_version[version] # : ::Hash[::String, untyped]
	end
	# Provides the set of available JSON schema versions.
	#
	# @return [Set<Integer>]
	# @see #json_schemas_for
	# @see #latest_json_schema_version
	#
	# @example Print the list of available JSON schema versions
	# artifacts = ElasticGraph::SchemaArtifacts::FromDisk.new(schema_artifacts_dir)
	# puts artifacts.available_json_schema_versions.sort.join(", ")
	def available_json_schema_versions
	@available_json_schema_versions ||= begin
	versioned_json_schemas_dir = ::File.join(artifacts_dir, JSON_SCHEMAS_BY_VERSION_DIRECTORY)
	if ::Dir.exist?(versioned_json_schemas_dir)
	::Dir.entries(versioned_json_schemas_dir).filter_map { |filename| filename[/v(\d+)\.yaml/, 1]&.to_i }.to_set
	else
	::Set.new
	end
	end
	end
	# Provides the latest JSON schema version.
	#
	# @return [Integer]
	# @raise [Errors::MissingSchemaArtifactError] when no JSON schemas files exist within the `artifacts_dir`.
	# @see #available_json_schema_versions
	# @see #json_schemas_for
	#
	# @example Print the latest JSON schema version
	# artifacts = ElasticGraph::SchemaArtifacts::FromDisk.new(schema_artifacts_dir)
	# puts artifacts.latest_json_schema_version
	def latest_json_schema_version
	@latest_json_schema_version ||= available_json_schema_versions.max || raise(
	Errors::MissingSchemaArtifactError,
	"The directory for versioned JSON schemas (#{::File.join(artifacts_dir, JSON_SCHEMAS_BY_VERSION_DIRECTORY)}) could not be found. " \
	"Either the schema artifacts haven't been dumped yet or the schema artifacts directory (#{artifacts_dir}) is misconfigured."
	)
	end

• ...whereas Results does not have them:

elasticgraph/elasticgraph-schema_definition/lib/elastic_graph/schema_definition/results.rb

Lines 32 to 61 in 2932af6

	def graphql_schema_string
	@graphql_schema_string ||= generate_sdl
	end
	# @return [Hash<String, Object>] the Elasticsearch/OpenSearch configuration dumped as `datastore_config.yaml`
	def datastore_config
	@datastore_config ||= generate_datastore_config
	end
	# @return [Hash<String, Object>] runtime metadata used by other parts of ElasticGraph and dumped as `runtime_metadata.yaml`
	def runtime_metadata
	@runtime_metadata ||= build_runtime_metadata
	end
	# @private
	STATIC_SCRIPT_REPO = Scripting::FileSystemRepository.new(::File.join(__dir__.to_s, "scripting", "scripts"))
	# @private
	def derived_indexing_type_names
	@derived_indexing_type_names ||= state
	.object_types_by_name
	.values
	.flat_map { |type| type.derived_indexed_types.map { |dit| dit.destination_type_ref.name } }
	.to_set
	end
	# @private
	def sourced_update_targets_by_source_type_name
	@sourced_update_targets_by_source_type_name ||= Indexing::SourcedFromUpdateTargetsResolver.new(state).resolve
	end

That's a mismatch that leads to .respond_to? checks which I'd like to avoid, and also compromises the clean "FromDisk and Results can are interchangeable story" we had before.

We should figure out a solution--maybe as part of this PR (or in another). (And we should make sure to follow the same pattern for proto ingestion).