feat: add multi-host HA connection support for PostgreSQL clusters

[Gowtham Raj Elangovan (gowtham500)]

Description

Add PGEngine.from_hosts() factory method that enables connections to high-availability PostgreSQL clusters with automatic failover, using psycopg3's native multi-host connection support.

Motivation

Production PostgreSQL deployments commonly use primary/replica topologies (e.g., Patroni, Citus, streaming replication) where clients need to connect to multiple hosts with automatic failover. Currently, PGEngine only supports single-host connection strings via from_connection_string() or from_engine(). This PR adds first-class support for multi-host HA connections with target_session_attrs routing, enabling users to direct queries to primaries, standbys, or any available node.

Changes

• New from_hosts() classmethod on PGEngine accepting multiple host addresses, per-host ports, and target_session_attrs for connection routing
• New _build_multi_host_connection_string() static method that constructs psycopg3-compatible multi-host connection URLs
• TargetSessionAttrs Literal type for valid routing options (any, read-write, read-only, primary, standby, prefer-standby)
• Input validation for empty hosts, port ranges (1-65535), host/port length mismatches, and comma-in-host injection
• URL encoding of user, password, and database via urllib.parse.quote_plus
• Single-port broadcast (one port applied to all hosts) and default port (5432) support
• Comprehensive unit tests for connection string building (validation, edge cases, all routing options)
• Integration tests for from_hosts() with table creation and engine kwargs passthrough

Checklist

• Code follows project style guidelines
• Self-review completed
• Documentation updated
• Tests added and passing
• No breaking changes

Example Usage

Connecting to an HA PostgreSQL cluster (primary + replicas)

from langchain_postgres import PGEngine, PGVectorStore
from langchain_core.embeddings import DeterministicFakeEmbedding

# HA cluster with 1 primary + 2 replicas (e.g., Patroni, Citus, streaming replication)
engine = PGEngine.from_hosts(
    hosts=[
        "pg-node1.example.com",
        "pg-node2.example.com",
        "pg-node3.example.com",
    ],
    user="app_user",
    password="secret",
    database="vectordb",
    ports=[5432],  # Same port for all nodes
    target_session_attrs="primary",  # Always route to the current primary
    pool_size=5,
    max_overflow=2,
)

# Use the engine as usual — failover is handled automatically by psycopg3
embedding = DeterministicFakeEmbedding(size=768)
engine.init_vectorstore_table("my_docs", vector_size=768)

store = PGVectorStore.create_sync(
    engine=engine,
    table_name="my_docs",
    embedding_service=embedding,
)

store.add_documents([...])