Shared file analysis

Computes file categories, line counts, aggregates, and hypothetical split costs.

The inventory and token-estimation commands share these helpers. The calculations use an approximate token model and do not observe actual model consumption.