docs: for_spectre_users.md says GPU benefits start at 1000+ nodes but actual auto-routing threshold is 500 nodes

[github-actions]

Problem

docs/for_spectre_users.md lines 109-111 state:

GPU acceleration — Large circuits (1000+ nodes) benefit significantly from GPU acceleration. Use engine.prepare(use_sparse=True) for circuits over ~1000 nodes.

The GPU auto-routing threshold is actually 500 nodes (not 1000+), as defined in vajax/analysis/gpu_backend.py:

gpu_threshold: int = 500  # line 22

This is confirmed in docs/performance_analysis.md:

GPU results for circuits below ~500 nodes reflect GPU kernel overhead on tiny workloads, not simulation inefficiency. The auto-threshold (gpu_threshold=500) prevents this in normal usage.

And in CLAUDE.md:

GPU threshold is 500 nodes (gpu_backend.py) — circuits below this auto-route to CPU

Expected Fix

Change 1000+ nodes to 500+ nodes in the GPU acceleration section of for_spectre_users.md.

Note: The same error exists in gpu_solver_architecture.md (tracked separately in issue #86).