feat(keywords): add estimated difficulty core

[hubab1]

Summary

• add a deterministic, versioned estimated keyword-difficulty heuristic over the top 10 ranking results
• return structured per-result evidence, confidence, unavailable reasons, algorithm identity, and explanatory notes
• preserve exact selected-result title/subtitle evidence and retain distinct apps when provider positions collide
• keep the core behavior-inert: no persistence, UI, MCP surface, network request, provider provenance, or Apple Ads metric state is changed here

Scoring model

• authority: position-weighted log10(ratingCount + 1) / 6, capped at one million ratings
• metadata saturation: keyword-token coverage in title and subtitle, with exact phrase matching retained as evidence only
• final score: 70% authority and 30% metadata saturation, rounded and clamped to 0...100
• confidence: geometric combination of top-10 sample coverage and rating-evidence coverage

The implementation explicitly identifies this as an OpenASO estimate, not Apple Ads difficulty, Search Popularity, or an Apple-provided metric.

Scope and follow-ups

This is the pure computation core for roadmap item 20a. Persistence and provenance belong in 20b; user-facing and MCP integration belong in 20c. This PR is stacked on #35 and remains draft while that manual provider-wiring gate is reviewed.

Attribution

The feature direction was informed by this upstream WIP commit: akshaynexus@5de98ea

The formula, evidence model, missing-data behavior, normalization, and tests were independently specified and rewritten for this repository. No co-author trailer is used.

Validation

• 20 focused estimator tests pass
• 283/283 complete tests pass
• signed Debug build and strict code-sign verification succeed
• exact built app remains running through a 3-second launch smoke test
• independent amendment review reports no P1/P2 findings
• prior architecture, formula, and attribution reviews report no P1/P2 findings