feat(memory): add Voyage AI embedding provider

[luongndcoder]

Summary

• Add VoyageEmbeddingProvider in internal/memory/voyage_embedding.go — a dedicated Voyage AI implementation that calls https://api.voyageai.com/v1/embeddings directly, with support for the input_type param ("query" / "document") for improved asymmetric retrieval quality
• Add ProviderVoyage = "voyage" constant and register it in ValidProviderTypes so users can create a Voyage provider via the API or UI
• Update buildEmbeddingProvider in cmd/gateway_agents.go to route the voyage provider type through VoyageEmbeddingProvider (Voyage is embedding-only; it has no chat API), with voyage-4 as the default model
• Add Voyage AI to the UI provider type selector (ui/web/src/constants/providers.ts) with the API base pre-filled

Motivation

Voyage AI produces state-of-the-art embeddings for retrieval, especially for code and domain-specific corpora. The existing OpenAIEmbeddingProvider cannot be used for Voyage because:

1. Voyage's input_type field (query vs document distinction) is not part of the OpenAI spec
2. Voyage has no chat endpoint — it should never be registered as a chat provider in the registry

This PR adds a thin, dedicated provider that satisfies the EmbeddingProvider interface and integrates cleanly with the existing embedding resolution flow.

Test plan

• Add a provider with type voyage and a valid Voyage AI API key via the UI
• Set embedding.provider in system configs to the provider name
• Send a message that triggers memory storage — verify embeddings are generated with model voyage-4
• Confirm input_type can be omitted (default) or configured per deployment