Consolidate/pr3 recall helpers

[jdye64]

Summary

Consolidates duplicated recall-evaluation helpers and pipeline utility functions that were copy-pasted across batch_pipeline.py, inprocess_pipeline.py, online_pipeline.py, and fused_pipeline.py into two shared modules.

• recall/core.py — Adds gold_to_doc_page() and hit_key_and_distance() alongside the existing recall functions. These were independently defined (with minor variations) in three pipeline scripts.
• examples/common.py (new) — Adds estimate_processed_pages() and print_pages_per_second(), previously duplicated in batch_pipeline.py and inprocess_pipeline.py.
• All four pipeline scripts are updated to import from the shared modules instead of defining their own copies.
• fused_pipeline.py now imports gold_to_doc_page, hit_key_and_distance, and is_hit_at_k from recall/core.py and estimate_processed_pages/print_pages_per_second from examples/common.py, instead of reaching into batch_pipeline's private functions.
• batch_pipeline._ensure_lancedb_table now uses lancedb_schema() from lancedb_utils (PR2) instead of an inline 10-field PyArrow schema definition.
• Removes a duplicate HF model registry entry in hf_model_registry.py.

Net diff: +134 / −227 lines (~93 lines removed)

Stacked on

• consolidate/pr2-lancedb-utils

Files changed

File	Change
nemo_retriever/src/nemo_retriever/recall/core.py	Added gold_to_doc_page() and hit_key_and_distance()
nemo_retriever/src/nemo_retriever/examples/common.py	New — estimate_processed_pages(), print_pages_per_second()
nemo_retriever/src/nemo_retriever/examples/batch_pipeline.py	Removed 5 local helper defs; uses lancedb_schema() in _ensure_lancedb_table
nemo_retriever/src/nemo_retriever/examples/inprocess_pipeline.py	Removed 5 local helper defs; imports from recall/core and examples/common
nemo_retriever/src/nemo_retriever/examples/online_pipeline.py	Removed 3 local helper defs; imports from recall/core
nemo_retriever/src/nemo_retriever/examples/fused_pipeline.py	Switches from batch_pipeline._* private imports to shared public APIs
nemo_retriever/src/nemo_retriever/utils/hf_model_registry.py	Removed duplicate nvidia/llama-nemotron-embed-1b-v2 entry
nemo_retriever/tests/test_create_local_embedder.py	Align expected alias to actual mapping
nemo_retriever/tests/test_lancedb_utils.py	CI stub adjustments from PR2 fixes

Test plan

• Existing unit tests pass (test_create_local_embedder, test_lancedb_utils, test_multimodal_embed)
• inprocess_pipeline, batch_pipeline, online_pipeline, and fused_pipeline execute end-to-end with recall evaluation enabled
• Verify gold_to_doc_page / hit_key_and_distance produce identical results to the removed per-pipeline versions
• Verify estimate_processed_pages / print_pages_per_second output matches previous behavior

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.