feat: pass IndexConfig to repositories with scalability options

[ochanism]

Overview

Pass VectorStore's IndexConfig to all repository constructors so that custom index/collection names and shard/replica settings take effect. Previously engine_factory ignored store.IndexConfig, causing all DB stores to use the same env-var defaults — breaking the core value of VectorStore separation.

Part of #921

Changes

IndexConfig struct (5 new scalability fields):

• ShardNumber, ReplicationFactor (Qdrant)
• ShardsNum, ReplicaNumber (Milvus — shards at CreateCollection, replicas at LoadCollection)
• DesiredShardCount (Weaviate)

Repository constructors (all accept optional *IndexConfig, nil = env fallback):

• ES v8: index name + shards/replicas in createIndexIfNotExists
• ES v7: index name + shards/replicas + new createIndexIfNotExists (was missing)
• Qdrant: collection prefix + shard/replication in ensureCollection
• Milvus: collection name + WithShardNum + WithReplica
• Weaviate: collection prefix + ShardingConfig + ReplicationConfig

Validation & security:

• ValidateIndexConfig: name regex + numeric bounds (shards ≤ 64, replicas ≤ 10)
• Auto-detect server version on CreateStore (prevents ES v7/v8 SDK mismatch)
• Sanitize ES v7 error messages to avoid cluster info disclosure

Other:

• Nil-safe getter helpers (pointer receivers) + ResolveIndexName/ResolveCollectionName with env fallback
• OptionalUint32 with negative overflow guard
• /types API updated with all scalability index_fields
• 30+ unit tests

Core

Key files:

• internal/types/vectorstore.go — IndexConfig fields, getters, validation
• internal/container/engine_factory.go — passes &store.IndexConfig to each repository
• internal/container/container.go — passes nil to env path (preserves existing behavior)
• Repository files: elasticsearch/v7/, elasticsearch/v8/, qdrant/, milvus/, weaviate/

Why

• Without this change, users can set index/collection names in the UI but they are silently ignored
• Scalability options (shards, replicas) are essential for production vector DB deployments
• Auto version detect prevents the wrong ES SDK from being used (v8 SDK → v7 server = 406 error)

Test Plan

• Getter helpers: nil receiver, zero, positive, negative values
• ResolveIndexName/ResolveCollectionName: nil fallback, env var, precedence
• OptionalUint32: zero, negative, positive
• ValidateIndexConfig: valid configs, name injection, numeric bounds (15 cases)
• Scalability fields JSON round-trip serialization
• Go build + go test + go vet clean
• Manual E2E: ES v7 index created with custom name + shards/replicas confirmed

Related

• Depends on: feat: add registry dual-map and engine factory for VectorStore #973 (PR3: Registry dual-map + Engine Factory)