feat: standalone claude-memory-mcp with multi-user support and Vault integration

Extracted from private infra repo into standalone open-source project.

Three operating modes:
- Local: SQLite + FTS5 (zero dependencies)
- Server: PostgreSQL via HTTP API with multi-user auth
- Full: PostgreSQL + HashiCorp Vault for secret management

Features:
- MCP stdio server with 5 tools (store/recall/list/delete/secret_get)
- FastAPI HTTP API with multi-user Bearer token auth (API_KEYS JSON map)
- Regex-based credential detection with auto-redaction
- AES-256-GCM encryption fallback for non-Vault deployments
- Vault KV v2 client (stdlib urllib, K8s SA auto-auth)
- Per-user data isolation (all queries scoped by user_id)
- Secret migration endpoint for existing plain-text credentials
- Backward-compatible env var aliases (CLAUDE_MEMORY_API_URL)

Infrastructure:
- Docker + docker-compose (API + PostgreSQL + optional Vault)
- Woodpecker CI (test → build → push → kubectl deploy)
- GitHub Actions CI (Python 3.11/3.12/3.13) + Release (GHCR + PyPI)
- Helm chart + raw Kubernetes manifests

96 tests passing across 6 test files.

src/claude_memory/api/models.py

@@ -0,0 +1,32 @@
+from typing import Optional
+
+from pydantic import BaseModel, Field
+
+
+class MemoryStore(BaseModel):
+    content: str
+    category: str = "facts"
+    tags: str = ""
+    expanded_keywords: str = ""
+    importance: float = Field(default=0.5, ge=0.0, le=1.0)
+    force_sensitive: bool = False
+
+
+class MemoryRecall(BaseModel):
+    context: str
+    expanded_query: str = ""
+    category: Optional[str] = None
+    sort_by: str = "importance"
+    limit: int = 10
+
+
+class MemoryResponse(BaseModel):
+    id: int
+    category: str
+    importance: float
+
+
+class SecretResponse(BaseModel):
+    id: int
+    content: str
+    source: str  # "vault", "encrypted", "plaintext"