perf: implement header stripping and serialization fast-paths

[suckserberg]

Summary

This pull request introduces core performance optimizations and protocol efficiency improvements. Specifically, it implements "header stripping" for fixed-size packets and adds specialized fast-paths for common array/map types.

Technical Implementation

• Header Stripping: Implemented a registry in Batcher that tracks fixed-size schemas. When a fixed-size packet is batched, the 2-byte size header is omitted, reducing overhead.
• Serialization Fast-Paths: Added optimized paths in Types for arrays of primitives (u8, f32, boolean, string8, string16). These paths bypass generic field calculations and use optimized buffer loops.
• Zero-Allocation Dispatch: Refactored Packet dispatch to use pcall(func, arg1) instead of creating anonymous closures, significantly reducing memory pressure under high-volume traffic.
• Bounds Check Optimization: Replaced manual Lua-level bounds checks with native buffer bounds checks for fixed-size fields, reducing instruction count on the hot path.

Changelog Entry

• [Core]: Implementation of header stripping for fixed-size packets and high-performance serialization fast-paths. Reduces protocol overhead and improves CPU efficiency for bulk data.

Type of Change

• Feature
• Bug Fix
• Performance Optimization
• Documentation Update
• Maintenance / Chore

Verification and Standards

• I have adhered to the Development Patterns.
• I have verified that all CI/CD linting and formatting checks pass.
• I have confirmed the changes within a Rojo-supported environment.
• I have updated the technical documentation for any API or architectural changes.

[protheeuz]

LGTM 👍🏻