Add benchmarks for HDC encoding throughput

[Tristan-Stoltz-ERC]

Summary

Measure and track HDC encoding performance across different input types (text, numeric, categorical). The encoding pipeline is in symthaea-core/src/hdc/.

What to benchmark

Add criterion benchmarks that track:

• Single word encoding time — currently measured at ~97ns warm
• Batch encoding throughput — ~379µs for a 10-word sentence (12% of cycle budget)
• Binding operation speed — XOR-based bind, cyclic permutation, bundle (majority)
• Similarity computation — Hamming distance on 16,384-dimensional binary hypervectors

Why this matters

HDC encoding is ~13.6% of the cognitive cycle. As we scale to longer inputs and more modalities, tracking regression is critical. The cognitive loop runs at ~234Hz (4.3ms/cycle) in release mode.

Getting started

1. Look at existing benchmarks in benches/ for patterns
2. The main encoding entry point is the TextEncoder in symthaea-core/src/hdc/
3. Use criterion with BenchmarkGroup for organized reporting
4. Run with cargo bench -p symthaea-core