Implement a simple token embedding lookup without using any deep learning library. Given a vocabulary size vocab_size, an embedding dimension embed_dim, and a list of integer token IDs token_ids, your function should:
(vocab_size, embed_dim) using numpy.random.default_rng(seed) with the provided seed, and the generator's standard_normal method..tolist()) where row i of the output is the embedding vector corresponding to token_ids[i].This is the foundational operation used at the input layer of any transformer-based language model: each token ID indexes into a learned (here, randomly initialized) embedding table to retrieve a dense vector representation.
Return the rows as a nested Python list (do not round inside the function; tests will round for display).