[fallintoplace]

What

Reduce allocations when encoding and decoding packed scalar extensions.

Why

Large packed fields can grow buffers repeatedly.

Implementation

• Reserve space for large fixed32 and fixed64 values.
• Preallocate buffers for packed scalar encoding.
• Cover partial tails, mixed records, and negative varints.
• Add opt-in Criterion benchmarks for decode and encode.

Benchmark

• Fixed64 decode, 4,096 values: 2.89 us to 2.24 us (22% lower).
• Fixed64 decode, 16,384 values: 20.16 us to 12.31 us (39% lower).
• Apple M1 Pro, Rust 1.95, Criterion, 50 samples. The numbers are for decoding.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}