feat: initial oracle-26ai-vector MCP server

MCP server exposing Oracle 23ai as a RAG vector store backed by
OCI GenAI Cohere Embed v4 (1536 dims). Three tools:
- insert_document: embed + store a text chunk
- update_document: re-embed + update an existing chunk
- search_similar: cosine similarity search (VECTOR_DISTANCE)

Uses python-oracledb thin mode with wallet (config_dir only).
Configured for Claude Code via ~/.claude/settings.json.

sql/schema.sql

@@ -0,0 +1,18 @@
+-- Oracle 23ai Vector Store Schema
+-- Run this against your Oracle 23ai instance before starting the MCP server.
+
+CREATE TABLE vector_store (
+    id          VARCHAR2(36)  DEFAULT SYS_GUID()   NOT NULL,
+    doc_id      VARCHAR2(255)                       NOT NULL,
+    chunk_text  CLOB                                NOT NULL,
+    embedding   VECTOR(1536, FLOAT32),
+    created_at  TIMESTAMP     DEFAULT SYSTIMESTAMP  NOT NULL,
+    updated_at  TIMESTAMP     DEFAULT SYSTIMESTAMP  NOT NULL,
+    CONSTRAINT pk_vector_store PRIMARY KEY (id)
+);
+
+-- Vector index for cosine similarity search
+CREATE VECTOR INDEX idx_vector_store_embedding
+    ON vector_store (embedding)
+    ORGANIZATION NEIGHBOR PARTITIONS
+    WITH DISTANCE COSINE;