v1

feat: neuroplastic bio-brains + hot-reload LLM + inhibitory gating

Replace hand-coded tier-3 reflex thresholds with real spiking neural networks:
LIF neurons, pair-wise STDP, reward-modulated (R-STDP) eligibility traces,
synaptic-scaling homeostasis, and per-module weight persistence under
%APPDATA%/Chimera/brains/. Each connectome (fly/worm/mouse) now learns user
patterns over time and survives restarts.

Wire LLM executive → SNN readout via a ``gate`` multiplier on the readout
threshold so policy changes actually "numb" or "sensitise" the bio-brain,
closing the architecture.md gap where inhibitory gating was stubbed.

Make ``LLMClientProxy`` hot-swap providers (Ollama↔OpenAI↔Anthropic) when the
onboarding wizard saves a new provider, so first-time users go from "wizard
done" → "live Jarvis" without a process restart. Align WormPolicy defaults
and FlyConnectome grid with the shipped config.yaml, and explicitly pin
app.tier3_reflex.neural as a PyInstaller hidden import.

Verified end-to-end: 58/58 tests pass, installer builds (ChimeraSetup.exe,
46 MB), frozen binary boots, serves dashboard on :8000, WebSocket live,
Ollama chat round-trips, mouse cortex reflex fires on real cursor motion.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>