Make custom JSON converters safe for multi-segment readers

[flobernd]

Summary

The hand-written DateTime, TimeSpan, Stringified, and fractional numeric converters read reader.ValueSpan directly to parse stringified numbers with Utf8Parser. ValueSpan is unsafe when Utf8JsonReader is constructed over a multi-segment ReadOnlySequence<byte> (it throws, or returns a non-contiguous span), so these converters only worked on the single-segment reader that JsonSerializer uses for stream deserialization. Any sequence-based deserializer, such as a streaming reader over a PipeReader, failed on the first stringified epoch date or duration. Surfaced while prototyping streaming deserialization for #8919.

Changes

• Add JsonReaderExtensions.GetValueBytes(ref reader, scoped Span<byte>): copies the current token's value into a caller-provided contiguous buffer (single-segment via ValueSpan, multi-segment by iterating ValueSequence segments), returns the length, and throws JsonException when the value exceeds the buffer.
• Update the 18 numeric/date converters to parse from buffer[..length], with the scratch buffer sized per type (bool/int 16, long/float 32, double and ISO date 64).
• IndexSettingsTimeSeriesConverter extended-year first-byte peek now reads reader.ValueSequence.First.Span / reader.ValueSpan directly, avoiding a GetString allocation.
• The scoped modifier on the buffer parameter is required so a stackalloc can be passed (otherwise the by-value Span would be escape-able through the unscoped ref Utf8JsonReader, CS8350).

Generated converters were already sequence-safe via the HasValueSequence-guarded ValueTextEquals, so no generator changes are needed. Builds clean across the target frameworks (verified net10.0 and netstandard2.0); the existing single-segment typed deserialize path is unchanged.

[github-actions]

💚 All backports created successfully

Status	Branch	Result
✅	8.19
✅	9.4

Questions ?

Please refer to the Backport tool documentation and see the Github Action logs for details