refactor(storage): remove the unused legacy Usage session scanner

[heeoneie]

Summary

SettingsStore.usageStats and packages/storage/src/usage-stats-store.ts have no production consumer. Desktop reads Usage through the Runtime Host: the settings:usageStats handler in runtime-host-usage-ipc-main.ts resolves via loadUsageStats(deps.client, ...), and the only .usageStats( call site in the repository is window.maka.settings.usageStats(range, host) at settings-surface.tsx:636, which goes through that same Host path. @maka/core/usage-stats/* is a separate module and is untouched, as are SettingsStore itself and projectDesktopUsageStats.

Removing the scanner also removes the only import that pulled node:sqlite into ./settings-store's module graph — usage-stats-store.ts imported createSqliteSessionMetadataStore. public-entrypoints.test.ts fails on only the declared entrypoints load node:sqlite until './settings-store' leaves SQLITE_BACKED_ENTRYPOINTS, so that list is updated in the same change with the reason, as its own comment asks. A consumer importing @maka/storage/settings-store no longer pays the SQLite load.

usage-stats-store is not in packages/storage's exports map, so this is not a public-API change.

Refs #4059

Verification

Ran:

• tsc -p packages/core/tsconfig.json and tsc -p packages/storage/tsconfig.json — clean.
• biome lint packages/storage and biome format packages/storage — clean. The format run is what caught the collapsed type import in fd35428.
• @maka/storage suite (node --test "dist/**/*.test.js"): 991 tests, 969 pass, 3 fail, 19 skipped.
• The same suite on unmodified main in the same container: 995 tests, 973 pass, and the same 3 failures — rejects a second authority for the same storage root in another process, does not classify a SQLite write failure as a migration blocker, and an unmarked read-only workspace fails without leaving marker state. Two of them assert that a permission denial rejects, and my test process runs as root, so chmod does not deny it. Unrelated to this change.
• Repository-wide grep for readUsageStats, usage-stats-store, and SettingsStore.usageStats: no remaining reference.

Not run: the Desktop build, Desktop tests, and e2e. The Electron postinstall and one registry dependency are blocked in my environment, so the Desktop side is verified by reference inspection only, not by execution. Worth a reviewer confirming npm --workspace @maka/desktop test.

AI use

Select exactly one:

• No generative tool made a substantive contribution
• Generative tooling made a substantive contribution

Tool(s) and scope: Claude Opus 5 — call-site verification, the edits themselves, and running the checks above. Each commit carries a Generated-by: Claude Opus 5 trailer.

Checklist

• Tests cover the change and fail without it
• Lint, format, typecheck and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes — described under Summary above
• No

[Astro-Han]

Approved at exact head e1f40e073609ca506fa773dd3721f08af1689e94.

Codex assisted with call-site tracing, package-boundary analysis, and exact-head verification. I independently reviewed the final diff and commit messages and own this approval.

I verified that Desktop Usage remains Runtime Host-owned; the removed Storage scanner has no production consumer or supported external contract; ./settings-store no longer reaches node:sqlite; and the change introduces no replacement state or parallel path. The exact-head test and windows_recovery checks are green, the branch is mergeable, and there are no unresolved review threads. No P0–P3 findings.