feat: add DecodeToUtf8 zero-alloc API, o200k benchmarks, and optimize Explore allocations

- Add DecodeToUtf8/GetDecodedUtf8ByteCount span API for zero-allocation decode into caller buffers
- Add o200k_base benchmark comparisons (Encode, CountTokens, Decode vs cl100k)
- Optimize Explore/ExploreUtfSafe with stackalloc to reduce allocations by 23%

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

Author: HavenDV