[GRV-28]: Publish llm-cost use-case and port reference docs

[riddim-developer-bot]

Summary

• Adds a developer-facing Use cases and extension ports section to both llm-cost-attribution and llm-cost-estimation READMEs, with use-case → user-value → ports tables that mirror each package's catalog.
• Includes synthetic, key-free library examples: an in-memory LinearEstimateSource, an in-memory EstimateTaggedUsageSource (and async-iterator variant), a flat-rate PricingTable, and a custom QuotaModel.
• Adds a Ready vs. planned APIs subsection in each README — attribution has none planned today; estimation lists calibrate as deprecated alongside its replacement calibrateCoverage.

Closes GRV-28.

Acceptance criteria

• Attribution README has a library-use section mapping major use cases (ForecastIssueCost, ForecastProjectCost, JoinCostWithFeature, ReadGitDiffs, CorrelateCostWithFeature) to user value and extension ports.
• Estimation README has a library-use section mapping enrichment / forecasting use cases (EnrichUsageWithEstimate, ForecastIssueCost, ForecastProjectCost, CalibrateCoverage) to user value and extension ports.
• Synthetic example using a custom in-memory usage source instead of filesystem transcripts (attribution: syntheticUsageRecords + async-iterator records() source; estimation: same syntheticUsageRecords fed through the re-exported forecaster).
• Synthetic example injecting a custom PricingTable (flat-rate) and QuotaModel (SLO-fraction) in attribution; flat-rate PricingTable in estimation.
• Synthetic example injecting a custom LinearEstimateSource (in-memory resolveEstimates) in estimation.
• Planned / deprecated APIs separated from ready APIs and not shown in ready-use examples (attribution: no planned surface; estimation: calibrate listed as Deprecated).
• Use-case and port names match docs/architecture/use-case-catalog.md and packages/llm-cost-estimation/docs/use-cases.md.
• Attribution claim "no telemetry pipeline, no database, no API keys" preserved at the top of the README and reinforced in the Ready/planned subsection (custom-port workflows are opt-in).

Out of scope per the issue: no runtime behavior changes, no npm publishing or version bumps.

Test plan

• npm test in packages/llm-cost-attribution — 200 / 200 pass; boundary check passes.
• npm test in packages/llm-cost-estimation — 37 / 37 pass, including the leak-safety guard (no-org-data.test.mjs) which scans every committed file in the package and now sees the new README content.

🤖 Generated with Claude Code