feat: add UTF-8 CountTokens, parallel encode, and span-optimized encode paths

Add CountTokens(ReadOnlySpan<byte>) for zero-alloc token counting from
UTF-8 bytes using stackalloc/ArrayPool. Add EncodeParallel for large text
encoding via parallel BPE processing. Optimize Encode(ReadOnlySpan<char>)
and EncodeWithAllDisallowedSpecial with dedicated span-based code path
using NET9 alternate dictionary lookups. Update benchmark results and add
CountTokensUtf8 and EncodeParallel benchmark categories.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: HavenDV