fix(desktop): name the session in the usage activity task column

[liuxiaocs7]

Summary

On the Usage Statistics page (使用统计) → Activity records (活动记录), the 任务 (Task) column showed 未命名会话 · <short id> ("Untitled session") for every row — even sessions that have a real title.

This is a regression. #3833 fixed exactly this column by carrying the session name (SessionHeader.name) through the old packages/storage/src/usage-stats-store.ts pipeline. #3697 then moved usage onto the Runtime Host projection and #4060 deleted that store. The new pipeline only carried sessionId, never the title, so UsageRequestLog.sessionName was declared-but-never-populated and the UI's usageSessionDisplayLabel always fell back to the untitled label.

Resolve the title on the Host, where every usage-bearing session is reachable:

• HostUsagePricingCoordinator takes a session-title reader (stores.sessionStore.readHeaderSnapshot). Per logs page it resolves the title for each row's sessionId and emits it as a new bounded sessionTitle field on the LLM/tool usage projections (projectText, ≤1024 bytes).
• Reading the durable session header by id bypasses the catalog's role/preparing/ledger-v0 filters (sqliteOrdinarySessionRolePredicate et al.), so reserved-role, coordination (e.g. WorkHub), and legacy sessions are named too — not just ordinary ones.
• Resolution is per-row and tolerant: a session the reader can't read is simply left untitled; one unreadable session never blanks the rest.
• The desktop layer copies sessionTitle → sessionName; the renderer (usageSessionDisplayLabel) already owns the untitled fallback, so no UI change.

Fixes #4218

Review responses

This supersedes the earlier main-process/catalog-join approach. Addressing the review:

• Spec P2 (catalog is not a complete usage-session source). Confirmed: readCatalogRecord applies the same role/preparing/ledger-v0 filters as the list query, so no complete name source is reachable from the desktop process. Fixed at the source — the Host resolves titles from the unfiltered durable header, covering reserved-role/coordination/legacy sessions.
• Spec P2 (one legacy session blanks all titles). Gone: resolution is per-row via readHeaderSnapshot, wrapped per-id, so a single unreadable/unprojectable session can't affect the others. The all-or-nothing client.listSessions() is no longer used here.
• Standards P2 (reconnect exception swallowed). Gone: the separate loadSessionNames() catch was removed. The title now rides the existing queryUsage read, which already propagates reconnectable failures through handleReconnectableRead.
• Standards P3 (blank-title test not actually covered). Covered on the Host: a coordinator test seeds a whitespace-only title and asserts the row is emitted without sessionTitle (removing the .trim() guard fails it), alongside a resolved title and a throwing (unreadable) session.

Verification

Same Storybook case (product-settings-pages--usage-long-tail), differing only by whether the title is populated. Named sessions show their real title; genuinely unnamed rows still fall back:

Checks run locally:

• tsc build of @maka/runtime-host and tsc -p tsconfig.main.json (desktop main) — both clean.
• node --test — 23/23 across the affected suites:
  • usage-pricing-two-client-uds incl. a new case: the LLM/tool logs carry the Host-resolved title, a whitespace-only title is dropped, and an unreadable session is tolerated.
  • usage-pricing-protocol incl. sessionTitle round-trip + over-length rejection.
  • runtime-host-usage-ipc-main: sessionName is surfaced from the projection's sessionTitle.
• biome check on the changed files — clean.

Not run in this environment: the full workspace typecheck / full test suite / e2e / a real Electron window. @maka/ui currently has pre-existing @astryxdesign/core type errors unrelated to this change; the desktop main process and the Host packages don't depend on @maka/ui, and Storybook renders it from source, so the evidence above is unaffected.

AI use

Select exactly one:

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

Tool(s) and scope: Claude Code (claude-opus-4-8) — diagnosis, the fix, the tests, and the before/after evidence. The commit carries a Generated-by 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