Split the proto_ingestion schema spec into cohesive files

[jwils]

Why

elasticgraph-proto_ingestion's schema_spec.rb had grown to 39 examples covering every proto schema generation concern, and "schema" is a broad enough label that it would keep growing. Per review feedback on #1306, the scalar-related examples belong in scalar_proto_types_spec.rb, and the other cohesive groups deserve their own files too.

What

Move each cohesive group of examples out of schema_spec.rb:

File	Covers	Examples
scalar_proto_types_spec.rb	how scalar types map to proto field types, including import: and comment:	9
stable_numbers_spec.rb	field and enum value numbers staying stable as the schema evolves	8
external_proto_enums_spec.rb	external_proto_enum, both sourcing values and referencing an external enum type	10
abstract_types_spec.rb	oneof wrappers for interface and union types	3
proto_syntax_and_headers_spec.rb	the syntax: and headers: options	3

schema_spec.rb keeps the 6 examples that cover schema generation as a whole, dropping from 1174 lines to 172.

Every example moved verbatim — I verified that the 39 example bodies are byte-identical before and after the split. schema_edge_cases_spec.rb is untouched.

Risk Assessment

None — spec file organization only. No behavior change, no coverage change (elasticgraph-proto_ingestion stays at 100% line and branch coverage).

References

• Addresses Map DateTime to google.protobuf.Timestamp and document temporal string formats #1306 (comment)
• Stacked on Use Buf to guard protobuf schema compatibility #1352 (→ Source enum values from external proto enums and reference external enum types #1286 → Support proto2 syntax and custom proto file headers #1305 → Map DateTime to google.protobuf.Timestamp and document temporal string formats #1306)

Note on the stack

#1286 and #1352 are currently based on an older copy of #1306's commit, so they need a restack. Rebase this branch after that restack. Expect a conflict in scalar_proto_types_spec.rb: #1306 now creates that file with its own examples, and this PR creates it with the moved ones, so the resolution is to keep both sets.