Add support for embedding generation

Summary

Implement embedding generation capabilities as outlined in docs/ARCHITECTURE.md and the development roadmap (#8, milestone 0.4.0).

Embeddings enable vector representations of text/content for use with vector databases, semantic search, and similarity comparisons.

Scope

Domain Layer

• Embedding DTO (src/Embeddings/DTO/Embedding.php) — vector validation, dimension helpers
• EmbeddingResult implementing ResultInterface — carries embeddings, token usage, provider metadata
• EmbeddingOperation for async/long-running embedding jobs

Builder & Client APIs

• PromptBuilder::withEmbeddingInput(string|Message $input) — collect inputs for embedding
• PromptBuilder::generateEmbeddingsResult() / generateEmbeddingsOperation() / generateEmbeddings()
• Matching methods on AiClient for traditional API surface
• Capability inference routes through CapabilityEnum::embeddingGeneration

Provider Contracts & Registry

• EmbeddingGenerationModelInterface
• EmbeddingGenerationOperationModelInterface
• WithEmbeddingOperationsInterface
• ProviderRegistry discovery for embedding-capable models

Provider Implementations

• OpenAI embedding model metadata & adapter
• Google / Anthropic stubs or implementations as available

Testing & CLI

• Unit tests for DTOs, builder flows, client methods, provider adapters
• Mock embedding model in test fixtures
• CLI support (--capability=embeddings, output formats)

Documentation

• README quick-start examples (fluent + traditional)
• Architecture/requirements doc updates

Related

• Roadmap: #8 (0.4.0 milestone)
• Architecture spec: docs/ARCHITECTURE.md

Open Questions

1. Input shape for multiple documents — new withEmbeddingInputs() vs. reusing message stack?
2. Helper transformers on EmbeddingResult (e.g., toVectors(), toAssociativeArray())?
3. Provider-specific options (dimensions) — surface through ModelConfig / SupportedOption?

Start with docs/ARCHITECTURE.md and the roadmap reference, then review the listed DTO, builder, client, provider-contract, registry, and provider paths. Resolve the open questions about input shape, result helpers, and provider-specific options before implementing the embedding capability. Done includes the scoped APIs, provider support, unit tests, and README and architecture updates.