[codex] cache things-cli read state

[pdurlej]

Summary

• add an incremental local state cache for things-cli read commands
• avoid the startup history.Sync() call for read-only commands
• fetch only delta items after the cached server index, rebuilding from zero only when the cache cursor is ahead of the server
• document THINGS_CLI_CACHE for overriding the cache file path
• add tests for command classification and cache round-tripping

Context

Before this change, every read command rebuilt memory.State from start-index=0, and initCLI() also fetched history items before dispatching the command. That made commands like list, show, areas, projects, and tags pay a full-history cost on every invocation.

The cache stores the aggregated memory.State plus the last server index. On subsequent read commands, the CLI checks the current server index and applies only new history items to the cached state. Write commands still sync the history head before writing so ancestor-index stays valid.

This branch is independent of the pagination fix in #6: before fetching a delta page, it seeds history.LoadedServerIndex with the cached cursor so the current main implementation continues from the correct position.

Validation

• go test ./cmd/things-cli -count=1
• go test ./...