feat(schema): distinguish input vs output JSON schemas (closes #2508)

[genisis0x]

Summary

Closes #2508.

Glaze previously emitted a single JSON Schema for every type even though serialization and parsing contracts differ. The mixed behavior — required on objects but no minItems on tuples after #2461 — described neither contract fully.

This PR introduces schema_purpose { input, output } and threads it through to_json_schema via an opts_with_schema_purpose wrapper that carries the tag in the existing constexpr Opts.

New entry points:

• glz::write_json_input_schema<T>() — partial-parse friendly: no required on objects, no minItems on tuples.
• glz::write_json_output_schema<T>() — serialization contract: required listed for guaranteed-present keys, minItems == maxItems on tuples.

glz::write_json_schema<T>() keeps its existing signature and now defaults to output purpose. For objects this preserves the existing required emission; for tuples this newly emits minItems == maxItems, completing the output-schema contract.

User-provided meta<V>::required is always honored regardless of purpose, since it expresses a developer-asserted parsing requirement. Variant discriminator tags also stay in required for input schemas because the tag must be present to select an alternative.

Test plan

• jsonschema_test builds cleanly.
• Updated tuple schema uses prefixItems expected string to include minItems.
• New schema_purpose_test suite covers:
  • input schema for tuples omits minItems,
  • output schema for tuples sets minItems == maxItems,
  • input schema for objects omits required (with error_on_missing_keys),
  • output schema for objects lists non-nullable members in required,
  • default write_json_schema matches output purpose.