v3.13.3

v3.13.3 — NIR Interop, Deep Audit, Docs Overhaul

NIR Bridge:
- 18/18 primitives (was 11), full roundtrip verified
- Cross-framework interop: Norse, snnTorch, SpikingJelly, Sinabs, Rockpool
- subtract-reset mode, auto-broadcast scalar params, threshold fix

Performance:
- Import time 200s → 10s (lazy-load neuron models + deferred scipy)

Deep Audit:
- 15 bugs + 7 concerns fixed (6-agent + Gemini audit)
- All claims verified by executing code
- Visualization plots 14→12 (actual count), HDL modules 10→16

Documentation:
- 38 tutorials (was 24), 8 notebooks (was 6), 19 examples
- 16 HDL modules listed, all API docs verified
- Anulum + Fortis Studio branding, dark/light toggle
- CHANGELOG, ROADMAP, ARCHITECTURE fully updated

Equation Compiler:
- ODE string → Python neuron + Q8.8 Verilog in one call
- Multi-variable ODEs (FitzHugh-Nagumo, Izhikevich)

SC Arithmetic:
- CORDIV division, adaptive bitstream length, Sobol/Halton decorrelation

2 155+ Python tests, 373 Rust tests, 100% coverage, all CI green.