New macros for nested struct relationship with automatic GTS segments generation

[Artifizer]

• Initial version [WIP] feat: implement schema inheritance and composition with schemars integration #21
• Naive detection of GTS instance and schema(type) fields in base GTS struct (check id/gts_id, type/gts_type)
• Advanced ideomatic version
  • Support minor version (when we have first out of process)
  • Check major/minor version in GTS string and struct suffix, e.g. V1
  • Prohibit scenario when non-GTS struct extends GTS struct
  • Tools must generate schema the same way as schemars
  • Use schemars in build time
  • allOf handling is unsafe if flattened/merged
  • Inheritance/nesting logic is brittle
  • Two independent generators can diverge
  • Compile-time constant schema is replaced with runtime generation
  • Detection of GTS instance and schema(type) fields in base GTS struct

[MikeFalcon77]

More details about addressed problems:

• allOf handling is unsafe if flattened/merged
  Any attempt to “collapse” allOf by merging only properties/required is not semantically equivalent to JSON Schema and can drop constraints (additionalProperties, combinators, numeric/string constraints, etc.), causing false-positive validation. resolve_schema_refs should inline $ref but preserve allOf structure unless a full normalization step is implemented and tested.

• Inheritance/nesting logic is brittle
  The proposed nesting path logic (collect_nesting_path / wrap_in_nesting_path) makes strong assumptions about generic chains and how schemas should be nested. It likely breaks for common Rust/serde patterns (Option<T>, Vec<T>, flatten, renamed fields). Needs clear spec + focused tests.

• Two independent generators can diverge

  • Proc-macro (gts-macros) generates GTS_SCHEMA_JSON at compile time.
  • CLI generator (gts-cli generate-from-rust) parses source via regex and uses its own type mapping.
  • These two paths can drift and emit different schemas for the same struct.

• Compile-time constant schema is replaced with runtime generation
  Current code produces GTS_SCHEMA_JSON: &'static str (embedded at compile time). The patch switches to methods that call schemars::schema_for!(Self) and serialize at runtime (allocating String). This is a behavior/performance/API shift and contradicts the “zero allocation / compile-time” positioning. If compile-time constants are required, we need a build-time generation approach (build.rs + include_str!) or keep the existing model.