JsonTextReader loses precision for double/decimal number tokens — no way to access raw source text

[sachinkale246]

When parsing JSON into a JToken, there is no way to preserve exact numeric precision for floating-point values that are representable by double or decimal but not both:

• double.MaxValue (~1.8e308) overflows decimal → FloatParseHandling.Decimal throws
• decimal.MinValue / high-precision decimals lose significant digits → FloatParseHandling.Double rounds
• There is no fallback mode that picks the best type per token
• JsonConverter.ReadJson cannot recover the raw source text because _stringReference is cleared before the token is emitted
  JSON and code related to the issue

[1.7976931348623157E+308, 1.23456789012345678901234567890]

// double.MaxValue overflows decimal → throws with FloatParseHandling.Decimal
// high-precision decimal loses digits with FloatParseHandling.Double (default)
var reader = new JsonTextReader(new StringReader(json));
reader.FloatParseHandling = FloatParseHandling.Double; // or Decimal — neither handles both
var token = JToken.ReadFrom(reader);

Suggested fix (with PR)
Two additions to JsonTextReader:

1. PreserveNumberText / NumberText — opt-in flag that captures the raw source text of each number token as a string? property, accessible after Read() returns. Zero cost when disabled.
2. protected virtual OnNumberParsed(string rawText, ref JsonToken tokenType, ref object value) — virtual hook called just before SetToken, allowing subclasses to replace the default parsed value with a more precise type (e.g. try decimal first, fall back to double) without reflection hacks.