Add Nemoretriever-parse (aka Eclair) NIM

[edknv]

Description

Part of https://github.com/NVIDIA/nv-ingest-private/issues/52
Also closes https://github.com/NVIDIA/nv-ingest-private/issues/49.

The NIM has its own docker-compose profile nemoretriever-parse so you have to specify --profile to start the NIM service, e.g.,

docker compose --profile nemoretriever-parse up --build

To submit a job via nv-ingest-client, specify extract_method.

CLI:

--task 'extract:{..., "extract_method": "nemoretriever_parse", ...}

Ingestor:

    .extract(
        ...
        extract_method="nemoretriever_parse",
        ...
    )

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[edknv]

321052f adds support for preserving the text bounding boxes in the metadata (in the hierarchy field) and addresses issue https://github.com/NVIDIA/nv-ingest-private/issues/49. It also changes the way text blocks are concatenated to concatenate text blocks with \n\n as requested by research team.