[akeit0]

Adds opt-in byte strings through LuaPlatform.DefaultByteStrings and LuaStringMode.Bytes: Lua strings carry bytes, file and stream I/O preserve raw data, and string classes/case conversion follow ASCII C-locale rules. UTF-16 remains the default, and existing main public APIs remain compatible.

The three original commits from #367 retain their authorship and hashes. This branch also includes the main fixes from #368.

• Keep C# source text, chunk names, paths and module-loader names as normal .NET strings. Convert at Lua boundaries; retain raw-byte loading for encoded source.
• Expose the string representation through LuaStringMode. Preserve existing constructors and stream-factory signatures; add mode-aware overloads.
• Use state.Load(module) for modules and share file-loading paths. Fix the recursive unbuffered Unity asset overload.
• Add mode-aware UnityStandardIO/DebugLogStream overloads so byte-mode UTF-8 text displays correctly in Unity logs. Preserve the existing constructors and UTF-16 defaults.
• Preserve raw I/O, CR handling, temporary-file mode and host-error conversion; avoid temporary encoding buffers in byte I/O.
• Encode UTF-8 directly into the byte-string char representation. Reuse ASCII strings and source spans; use one pooled char buffer for non-ASCII source loading. Preserve BOM handling, binary-chunk rejection and .NET surrogate replacement behavior.
• Run the 19 enabled official Lua scripts in both modes. Preserve the original files.lua bytes and byte-mode UTF-8 BOM checks; keep UTF-16 adaptations in that mode's harness. Vendored scripts are unchanged.

Validation on Windows, .NET 8.0.32:

• 713 runtime tests passed, excluding the existing ExpectedFailure category, including 38 official-script runs. Existing unsupported reader loading, bytecode dumping and process execution remain excluded by the harness.
• 69 source-generator tests and 76 focused tests against the netstandard2.1 assembly passed.
• All four library targets build with zero warnings/errors; formatting and public API compatibility checks pass. A client compiled against main runs with the preserved constructors.
• The converter oracle checks all UTF-16 code units, sampled supplementary scalars, malformed surrogates and ASCII-copy boundaries against Encoding.UTF8.
• BenchmarkDotNet, .NET 8 x64, 4K chars: standalone conversion time falls 94% for ASCII, 12-13% for sparse Unicode/Japanese and 3% for mixed text. ASCII conversion allocates nothing; other inputs allocate about one-third less. Exploratory whole-script loading improves about 8% for ASCII, while ordinary Unicode gains are small. The reproducible converter benchmark is included.
• Unity 6000.5.5f1 on Windows: 11/11 checks passed in Editor Mono and 11/11 in a native IL2CPP player; the native build succeeded and the player exited 0. Coverage includes real Lua assets, buffered/unbuffered execution, Unicode Resources modules and filenames, binary I/O, converter surrogate/ASCII-copy boundaries, and split UTF-8 writes to normal/error logs.

Draft pending maintainer approval. Use a merge commit when approved to preserve the contributor's commits and indirect PR merge status.