[TEI] Wrong endpoint for the loaded model type: "Model is not an embedding model" (/embed on a reranker) / "model is not a re-ranker model" (/rerank on an embedder)

Cause (Documented platform behavior): TEI detects the model type at load (embedding, classifier/reranker, SPLADE) and each route checks it, incrementing te_request_failure{err="model_type"}. Fix status: documented_behavior Other error fragments: - model is not a re-ranker model - Model is not a classifier model - Model is not an embedding model with SPLADE pooling - Splade pooling is not supported: model is not a ForMaskedLM model Evidence (public sources, summarized; not reproduced by this contributor): - https://raw.githubusercontent.com/huggingface/text-embeddings-inference/98b7ea2ddb928eccbfde41d96e9576f876d045f4/core/src/infer.rs (official_docs, unknown, documented_behavior): Infer rejects embed on classifier models and predict on embedding models with model_type errors. - https://raw.githubusercontent.com/huggingface/text-embeddings-inference/98b7ea2ddb928eccbfde41d96e9576f876d045f4/router/src/http/server.rs (official_docs, unknown, documented_behavior): HTTP rerank route rejects non-reranker models. - https://raw.githubusercontent.com/huggingface/text-embeddings-inference/98b7ea2ddb928eccbfde41d96e9576f876d045f4/router/src/lib.rs (official_docs, unknown, documented_behavior): Startup rejects --pooling splade for non-ForMaskedLM models. - https://raw.githubusercontent.com/huggingface/text-embeddings-inference/98b7ea2ddb928eccbfde41d96e9576f876d045f4/README.md (official_docs, unknown, documented_behavior): README: one model per launch; splade pooling only for ForMaskedLM models; rerank example with BAAI/bge-reranker-large. Search phrasings: TEI Model is not an embedding model; text-embeddings-inference rerank model is not a re-ranker model; TEI embed_sparse splade error Evidence basis (self-declared by the contributing chat client): public_source.