TokenStep 0.1.48: reduce background energy usage - #11

What changed

• Add power-aware refresh scheduling: at least 15 minutes on AC and 30 minutes on battery or Low Power Mode.
• Decouple usage collection, quota refresh, and Agent Work Rank refresh with independent TTLs.
• Replace repeated full Codex cache rewrites with a per-session SQLite incremental cache and unchanged-source checkpoint.
• Add safe cache migration, corruption recovery, periodic strong validation, and fork-anchor performance improvements.
• Bump the app and packaging defaults to v0.1.48.

Why

TokenStep previously launched a relatively expensive collection subprocess every five minutes and repeatedly encoded and rewrote a large collector cache even when local usage had not changed. On large Codex histories this created visible periodic CPU, memory, disk-write, and battery pressure.

User impact

• Background refresh uses substantially less CPU and disk I/O.
• Foreground panels still refresh according to the configured freshness interval.
• Token totals, model splits, daily distribution, and Agent activity accounting remain unchanged.
• Existing caches migrate automatically; corrupt incremental state is rebuilt safely.

Validation

• Codex cumulative, append, fork, rewrite, migration, and corruption-recovery fixtures passed.
• CC Switch and usage recalibration migration fixtures passed.
• Localization and language-refresh checks passed.
• Full TokenStep app and helper compilation passed.
• Frozen 7.1 GB history: 1,010 sessions / 62,392 records, zero accounting mismatches against a full rebuild.
• Unchanged warm refresh: 98 ms, 24 MB peak RSS.
• Active-session incremental refresh: 210 ms, 27 MB peak RSS, about 0.70 MB logical writes.
• Full XCTest execution is left to GitHub Actions because the local CommandLineTools installation does not provide the XCTest module.