taskToArchiveEntry() builds the archived record from an explicit field
whitelist that omits task.tokenUsage, and ArchivedTaskEntry has no
tokenUsage field — so a task's input/output/cache token (and derived cost)
accounting is silently dropped the moment it is swept out of the live
tasks table into archive.db. Every archived task loses its token stats.
Add tokenUsage to the ArchivedTaskEntry type and thread it through both the
archive write path (taskToArchiveEntry) and the two restore paths
(archiveEntryToTask, unarchiveTask) so the counts round-trip intact.
cdf83eda45