Added akms_nodes_gen.nlm_batch, a serial NotebookLM batch generator that
runs one selected batch/cluster against that batch's notebook.
Added CLI parameters for source IDs, timeout, output format, NotebookLM
prompt file, and output template file so generation policy stays outside the
Python implementation.
Added local mitigation gates for NLM-only generation: response caching,
raw-response audit files, resumable state, fenced YAML/JSON parsing, schema
checks, known-edge validation, required source references, and repair retries.
Added focused unit tests for batch loading, structured response parsing,
grounding/edge validation, and serial run behavior.