Align Holmes NLP component coverage

[mdheller]

Summary

Adds the first explicit Holmes NLP component alignment contract and expands the Holmes surface validator so the product spine covers lower-layer NLP families instead of only high-level investigation components.

This PR adds:

• docs/NLP_COMPONENT_ALIGNMENT.md
• explicit componentFamilies in examples/holmes-surface.json
• explicit nlpTasks coverage for primitive, advanced, rule, classical ML, neural, transformer, task-model, graph, and governance lanes
• stronger requiredPromotionEvidence fields for algorithm family, task contract, and latency/footprint evidence
• validator checks for component families and NLP task coverage
• README pointer to the alignment contract

Why

Holmes should absorb a disciplined NLP component taxonomy while remaining broader than a component NLP library. The target architecture is not a model zoo: it is a governed investigation fabric spanning primitives, extraction, classification, retrieval, evidence, graph conversion, policy, evals, and agent/tool orchestration.

Boundary

This PR makes the architecture enforceable at the Holmes contract layer. Runtime implementation and benchmarks belong next in SociOS-Linux/nlplab, with stable service graduation into SocioProphet/prophet-platform and pointer-backed retrieval through SocioProphet/sherlock-search.

Validation

Expected local validation:

make validate

The validator now requires:

• product components
• component families
• NLP task coverage
• promotion evidence fields
• cross-repo integrations