JSON parser DoS prevention

As reported by Claude review, our current JSON parser does not prevent large exponents from exhausting our memory.
The solution proposed by Claude, however, is inflexible: cap the exponent.
For most regular uses that would be OK. For schema validation, though, that might be too restrictive.
Instead, I suggest we just keep the exponent as a separate field for JSON numbers.
The trade off:

• We get arbitrarily large numbers with efficient-ish representation
• We allow users to define schemas with arbitrary numbers

• We need to implement basic operations and might need some float-like behavior in places

El issue no menciona archivos, pruebas ni puntos de entrada. Empieza por localizar el analizador JSON y su representación de números; después, determina el alcance de conservar los exponentes y las operaciones básicas necesarias; se considerará terminado cuando los exponentes grandes ya no provoquen agotamiento de memoria y la validación del esquema siga aceptando números arbitrarios.