Performance

PyInj targets O(1) resolution and predictable overhead:

• O(1) type lookups via token tables
• Cached signatures to avoid repeated reflection
• Lock-free fast path for singletons
• Minimal per-binding memory footprint

# Pseudo-benchmark
# 1000 services registered
# ~0.0001ms resolution, ~500 bytes/service