v0.2.0

Synced Leia's knowledge with current astromesh (core v0.28.5, Nexus v0.3.0). This makes the architect generate correct, current manifests and the operator speak the new control-plane API.

Fixed

• Templates: memory config was silently ignored. All 6 templates used a flat memory: block, but the runtime reads memory nested under conversational (astromesh/core/memory.py). Converted every template to spec.memory.conversational.{backend,strategy,max_turns} so the configured window actually applies.
• Templates: max_length guardrail was silently ignored. The runtime reads max_chars (astromesh/core/guardrails.py), not limit. Renamed the field in all 6 templates so the response cap is enforced.
• leia-architect emitted invalid patterns and sections. It told the architect to use single/chain/router/parallel (none of which exist) and to write spec.system_prompt/spec.constraints. Corrected to the real patterns (react, plan_and_execute, pipeline, parallel_fan_out, supervisor, swarm) and real sections (spec.prompts.system, spec.guardrails, nested spec.memory).

Changed

• schemas/astromesh-v1-agent.md modernized to astromesh v0.28.x: added spec.model.extra (multi-provider ranking), more generation parameters (top_k, frequency_penalty, presence_penalty, stop), round_robin routing (default is cost_optimized), MCP/webhook/rag tool types, postgres conversational + pgvector/qdrant semantic memory backends, corrected guardrail fields (pii_detection action redact, content_filter blocked_patterns, cost_limit max_tokens_per_turn), the {{contact_name}} prompt variable, and a note that channels are wired via env vars + per-agent webhook (no spec.channels).
• schemas/nexus-api.md updated to Nexus v0.3.0: dual auth (JWT Bearer or X-API-Key), new /auth/register|login|refresh, /api/v1/me, tenant management, and per-tenant API-key management; agent endpoints now accept either credential.
• schemas/whatsapp-config.md: documented the per-agent webhook path (/v1/agents/{name}/channels/whatsapp/webhook), the contact_name context variable, and system-direction delivery receipts.
• Provider guidance: clarified that only ollama/openai/openai_compat/azure_openai are runtime-wired; reach vLLM/llama.cpp/TGI via openai_compat.

Full Changelog: v0.1.0...v0.2.0