v0.95.0

Added

• FluxIndex takes a Microsoft.Extensions.AI embedding generator directly. FluxIndexContext.CreateBuilder() .UseEmbeddingGenerator(documents, queries) — or services.AddEmbeddingGeneratorEmbedding(queriesServiceKey) over the
  container's generator — embeds with any IEmbeddingGenerator<string, Embedding<float>> (OpenAI, Azure OpenAI, Ollama,
  LMSupply, IronHive) without an IEmbeddingService implementation. An asymmetric model's query side is a second
  generator; without it the documents generator embeds queries. Batches are one generator call. The index identity is the
  generator's provider, model id and dimension: a generator that reports no model id, or no dimension when
  EmbeddingGeneratorServiceOptions.Dimensions is unset, throws at construction, and a vector of another size throws
  when it arrives. FluxIndex.Core now references Microsoft.Extensions.AI.Abstractions.

Removed

• Breaking: IEmbeddingService.GetMaxTokens and CountTokensAsync (and their EmbeddingServiceBase defaults) —
  nothing in FluxIndex read them. Migration: delete those members from your implementations; an override of either no
  longer compiles.

Changed

• Breaking: the enrichment pipeline's EmbeddingGenerationOptions record is now MultiRepresentationEmbeddingOptions
  (IDocumentEnrichmentPipeline.GenerateEmbeddingsAsync), so it no longer collides with Microsoft.Extensions.AI's type
  of that name in files importing both. Migration: rename the type where you construct it.