feat: Intelligence Pipeline v3 — full implementation

Multi-stage evidence-grounded inference architecture replacing the
monolithic 9B model extraction pipeline. CPU-first specialist services
handle routine extraction while the 9B vLLM model is preserved for
semantic adjudication of ambiguous cases.

Key components:
- Capability-aware inference gateway (OpenAI-compatible + Ollama)
- Endpoint registry with DB migrations and REST API
- Sentence-aware document segmenter (property tests)
- Deterministic financial parsing with offset integrity
- Symbol resolution with ambiguity detection
- Specialist service (GLiNER2, dynamic batching, K8s deployment)
- Company-specific sentiment (FinBERT, calibration)
- Retrieval-based novelty and duplicate detection
- Confidence calibration pipeline
- Deterministic routing engine (property tests)
- 9B adjudication layer with VRAM gating
- Stock-specific impact model (features, labels, baseline, trained)
- Pipeline orchestrator (state machine, queues, leases, feature flags)
- Bounded parallelism (async workers, semaphore, load shedding)
- Observability (tracing, metrics, alerts)
- Compatibility adapter (v3→v2 golden mapping tests)
- Shadow/canary promotion framework
- Active learning and fine-tuning pipeline

Test results: 1,161 tests pass, ruff lint clean.
All 282 spec tasks completed.

infra/migrations/042_seed_inference_registry.sql

@@ -0,0 +1,74 @@
+-- Migration 042: Seed Inference Registry
+-- Populates initial endpoint profiles and model deployments for the
+-- existing Ollama and vLLM services.
+--
+-- Task 18.1: Create the current Ollama endpoint profile
+-- Task 18.2: Create the current vLLM OpenAI-compatible endpoint profile
+-- Task 18.3: Create model deployments matching actual runtime state
+--
+-- This is a DATA migration. The schema was created in 040_inference_registry.sql.
+-- Uses ON CONFLICT DO NOTHING for idempotency.
+
+-- ─── 18.1: Ollama endpoint profile ───────────────────────────────────────────
+INSERT INTO inference_endpoints (id, name, protocol, base_url, auth_secret_ref, auth_scheme, default_headers, health_path, enabled)
+VALUES (
+    'a0000000-0000-4000-8000-000000000001'::uuid,
+    'stonks-ollama',
+    'ollama_native',
+    'http://ollama.ollama-service.svc.cluster.local:11434',
+    NULL,
+    'none',
+    '{}',
+    '/api/tags',
+    TRUE
+)
+ON CONFLICT (name) DO NOTHING;
+
+-- ─── 18.2: vLLM OpenAI-compatible endpoint profile ──────────────────────────
+INSERT INTO inference_endpoints (id, name, protocol, base_url, auth_secret_ref, auth_scheme, default_headers, health_path, enabled)
+VALUES (
+    'a0000000-0000-4000-8000-000000000002'::uuid,
+    'stonks-vllm',
+    'openai_chat',
+    'http://kube-vllm.stonks-oracle.svc.cluster.local:8000',
+    NULL,
+    'none',
+    '{}',
+    '/health',
+    TRUE
+)
+ON CONFLICT (name) DO NOTHING;
+
+-- ─── 18.3: Model deployments ─────────────────────────────────────────────────
+
+-- Ollama model deployment (qwen3.5:9b served via Ollama native protocol)
+INSERT INTO model_deployments (id, endpoint_id, served_model_name, display_name, capabilities, context_window, max_output_tokens, quantization, runtime_metadata, enabled)
+VALUES (
+    'b0000000-0000-4000-8000-000000000001'::uuid,
+    'a0000000-0000-4000-8000-000000000001'::uuid,
+    'qwen3.5:9b',
+    'Qwen 3.5 9B (Ollama)',
+    '{"chat_completions": true, "json_schema": false, "json_object": true, "seed": false, "usage": false, "max_completion_tokens": false, "model_listing": true}',
+    32768,
+    32768,
+    NULL,
+    '{"source": "ollama_native", "notes": "Ollama-served model with native JSON mode"}',
+    TRUE
+)
+ON CONFLICT (endpoint_id, served_model_name) DO NOTHING;
+
+-- vLLM model deployment (AxionML/Qwen3.5-9B-NVFP4 on RTX 4070 Ti SUPER)
+INSERT INTO model_deployments (id, endpoint_id, served_model_name, display_name, capabilities, context_window, max_output_tokens, quantization, runtime_metadata, enabled)
+VALUES (
+    'b0000000-0000-4000-8000-000000000002'::uuid,
+    'a0000000-0000-4000-8000-000000000002'::uuid,
+    'AxionML/Qwen3.5-9B-NVFP4',
+    'Qwen 3.5 9B NVFP4 (vLLM)',
+    '{"chat_completions": true, "json_schema": true, "json_object": true, "seed": true, "usage": true, "max_completion_tokens": true, "model_listing": true}',
+    8192,
+    2048,
+    'NVFP4',
+    '{"gpu": "RTX 4070 Ti SUPER", "gpu_memory_utilization": 0.80, "max_num_seqs": 8, "vllm_structured_outputs": true}',
+    TRUE
+)
+ON CONFLICT (endpoint_id, served_model_name) DO NOTHING;