Lazy import llama_index in milvus utility to avoid unwanted side effects

[edknv]

Description

This PR moves from llama_index.embeddings.nvidia import NVIDIAEmbedding inside the nvingest_retrieval() function in order to prevent top-level import from triggering logging:

[nltk_data] Downloading package punkt_tab to
[nltk_data]     /opt/conda/envs/nv_ingest_runtime/lib/python3.10/site-
[nltk_data]     packages/llama_index/core/_static/nltk_cache...
[nltk_data]   Unzipping tokenizers/punkt_tab.zip.

and downloading when the module is not used.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.
• If adjusting docker-compose.yaml environment variables have you ensured those are mimicked in the Helm values.yaml file.

[jperez999]

Hopefully the import time does not cause a problem (notable decrease in qps) for a user who is trying to use the nvingest_retrieval function in a loop over lots of requests. Until we get a complaint, I am fine with this move.