Add DetectFromBytes(ReadOnlySpan<byte>) overload with comprehensive tests

[Copilot]

Applies PR #204 changes to add a ReadOnlySpan<byte> overload for DetectFromBytes, allowing callers to detect encoding from a span or slice without copying to a byte[]. Also addresses review feedback and bugs found during integration.

Source changes

• New overload: CharsetDetector.DetectFromBytes(ReadOnlySpan<byte>) — existing byte[] overloads now delegate to it
• Internal signature updates: Feed, HandleData, FindInputState, IsStartsWithBom, FindCharSetByBom, RunProber, all probers, and analyzers updated to accept ReadOnlySpan<byte> instead of byte[] + offset/len
• WriteSpanToStream compile guard: changed from NETSTANDARD2_1_OR_GREATER || NETCOREAPP2_1_OR_GREATER → NET8_0_OR_GREATER per review comment (targets are netstandard2.0 and net8.0)
• JapaneseContextAnalyser.HandleData: removed stale offset/len parameters, now span-only for consistency
• Empty span guard: added early return in Feed() when buf.Length == 0; several probers (e.g. EUCKRProber, SJISProber) unconditionally access buf[max - 1] and would throw IndexOutOfRangeException on empty input

Tests

Added 16 new tests covering the ReadOnlySpan<byte> overload:

// Empty span — should return no detection, not throw
CharsetDetector.DetectFromBytes(ReadOnlySpan<byte>.Empty); // Detected == null

// Key use case: detect from a slice without allocating a copy
ReadOnlySpan<byte> slice = largeBuffer.AsSpan(offset, len);
var result = CharsetDetector.DetectFromBytes(slice);

Covers: empty span, ASCII, UTF-8 (with and without BOM), sliced span, UTF-7/GB18030/UTF-16 BE+LE/UTF-32 BE+LE BOM detection.