Reduce per-byte overhead in VLQ integer decoding

## Description

The current VLQ decoder calls `get_aligned` for each byte, which involves repeated offset calculations and bounds checks in the hot loop.

## Proposed Change

Align to the byte boundary once, then iterate directly over the buffer slice, avoiding per-byte overhead from `get_aligned`.

🤖 Generated with [Claude Code](https://claude.com/claude-code)