Reduce allocation overhead in file type detection

[Copilot]

File type detection was allocating more than necessary on hot paths, especially for stream inputs and text/UTF-8 validation. The result was avoidable heap pressure during repeated inspections.

• Stream buffering

  • Rent buffers from ArrayPool<byte> instead of allocating a full byte array per stream detect call.
  • Preserve seek position while reading and return pooled buffers promptly.

• Matching path

  • Remove the temporary List<int> used for subheader scans.
  • Scan subheader candidates in-place against the current span.

• Text detection

  • Cache UTF-8 encodings and BOM bytes once.
  • Remove per-call encoding construction and LINQ-based BOM checks.
  • Keep validation span-based to avoid extra intermediate allocations.

ReadOnlySpan<byte> input = fileContent.AsSpan(0, length);
if (IsMatchingType(input, fileTypeInfo))
    return fileTypeInfo;

[ghost1face]

Thanks for the feedback, will pull into #10 and add benchmarks and minor changes