Preserve typed service names and paths in one schema

[KidkArolis]

What changed

• preserve both the literal schema name and literal transport path on every materialized Service
• make service<T>() return an unbound declaration whose path defaults to its eventual schema key
• add service<T>().at('api/path') for handwritten aliases without repeating the path in a generic and a value
• add service.from<ApiSchemaTypes>() for generated, path-keyed service catalogs, where the selected path determines both the runtime path and service contract
• model default-path and path-bound declarations as separate states, without nullable path generics or constructor casts
• make createSchema() the sole materialization boundary and reject empty or duplicate transport paths
• keep name and path namespaces explicit: builders, relationships, and current mutation APIs use schema names; descriptors, raw Feathers access, and deprecated hooks use transport paths
• expose explicit ServiceByName, ServiceByPath, and ServiceDefinitionByPath utilities instead of an ambiguous name-or-path lookup
• retain support for a schema name matching another service's transport path because the calling API determines the namespace
• make schema-bound useFeathers() return the actual underlying client with path-derived types, without a pass-through proxy

API

Handwritten schemas:

const schema = createSchema({
  services: {
    tasks: service<TaskService>(),
    people: service<PersonService>().at('api/people'),
  },
})

Generated path catalogs:

const apiService = service.from<ApiSchemaTypes>()

const schema = createSchema({
  services: {
    people: apiService('api/people'),
    companies: apiService('api/companies'),
  },
})

The object key supplies the Figbird name. The explicit path or catalog selection supplies the transport path. createSchema() produces one materialized schema carrying both.

Why

Applications migrating gradually to named Figbird APIs should not need a parallel "legacy schema" for path-based hooks. The previous PR iteration preserved paths, but declaring one required repeating the same literal in a generic and at runtime, and the generic ServiceByIdentifier utility became ambiguous when the name and path namespaces overlapped.

This version makes the two namespaces first-class and deliberately separate while keeping a single source of service types. It also gives schema generators a compact, checked API: unknown catalog paths fail at compile time and automatically select the matching backend contract.

Validation

• npm run tsc
• npm run lint
• npm run build
• npx ava test/type-inference.test.ts test/schema.test.tsx — 12 tests passed
• npm test — type-checking, lint, formatting, coverage, and all 359 tests passed