Harden architecture, durability, and long-history workflows

[wellivea1]

Summary

• add the Phase 4 desktop-local assistant and transactional installer/update path
• harden sleep/task sync with bounded batches, typed tombstones, atomic task transitions, transactional pull pages, and resurrection prevention
• replace long-history hot paths with indexed/paginated storage, high-water proposal cursors, linear estimation/scheduling, coalesced refreshes, and lazy frontend routes
• make sleep import authorization one-use and stale-safe, and make exports durable through same-directory atomic replacement
• add durable Android SQLite state, Health Connect paging, temporal/source-revision semantics, retry behavior, and a desktop-consistent dense UI
• unify sleep-v1 classification and contract/fixture validation across core, server, desktop, tools, and CI
• record all 33 review findings, dispositions, residual risks, and roadmap follow-up in docs/architecture-performance-review-2026-07-27.md

Validation

• go test ./core/... ./apps/server/... ./apps/desktop/...
• scoped go vet plus tools-module tests/vet
• contract validation and exact generation of 29 synthetic fixtures
• installer suite: 45 passed, 0 failed; modified PowerShell scripts parse cleanly
• desktop frontend: 243 tests; trusted frontend: 6 tests; typecheck, ESLint, Prettier, UI standards, and production builds pass
• Android JVM tests rerun from scratch; lintDebug and assembleDebug pass
• Android instrumentation: 4/4 on Pixel_10_Pro_XL_API_36_1; Status, Correct sleep, and Medication event visually inspected
• desktop, MCP, and server binaries compile
• backtest and scheduler benchmarks cover 100 through 20,000 records
• git diff --check, credential-pattern scan, and git fsck --full pass

Review scope

The review ledger distinguishes completed, partial, and intentionally deferred findings. Remaining architectural debt is explicit in the ledger and roadmap rather than being represented as complete.