LLM 0.32a1 patches tool-call conversation reinflation bug from SQLite

Simon Willison shipped a quick follow-up release to his LLM command-line tool, tagged 0.32a1. The single fix addresses a regression introduced in the prior 0.32a0 alpha where tool-calling conversations failed to rehydrate correctly when read back from the SQLite store that LLM uses for persistent history.

The bug, tracked as issue #1426, broke continuity for any workflow that relied on resuming agentic sessions across invocations — a core use case as LLM has expanded into multi-turn tool execution. The point release keeps the larger 0.32 refactor on track without forcing users to choose between the new architecture and a working tool-call replay path.