Lossless numeric option for `JsonValue` parsing to preserve `Decimal`

[CasperKristiansson]

When using JsonValue::parse_with_config, fractional JSON numbers are parsed into Float(f64), so the original literal is lost. Downstream (e.g., pydantic-core’s JsonValidator path) cannot build a Decimal from the original text and only sees f64.to_string(), causing precision loss.

Ask: add an opt‑in mode on the JsonValue parsing path that preserves the raw numeric token (e.g., a lossless number variant or raw text) so consumers can construct Decimal without going through f64.

References:

• crates/jiter/src/value.rs: JsonValue enum currently uses Float(f64) for fractional numbers.
• pydantic-core invokes JsonValue::parse_with_config in pydantic-core/src/validators/json.rs::JsonValidator.
• Related pydantic issue: Decoding Decimal from JSON number is lossy pydantic#9180 (Decimal precision loss when parsing JSON numbers via JsonValue).

Context: Workarounds are parsing via PythonParse with FloatMode::Decimal or passing numbers as strings. An opt‑in lossless JsonValue mode would let pydantic-core keep its existing pipeline while preserving precision.

[davidhewitt]

Seems reasonable to me, also relates to #232