Gathering kindle formats right before extracting

[yangyubo]

I use Calibre on a 2GB RAM Synology NAS machine, the plugin consumed 800MB (or more) extra RAM while was converting hundreds of large AZW3 books. So before the plugin finishing the task, Docker killed Calibre for memory exhausted.

This patch postpones kindle formats gathering, the def multi_dispatcher() no need to hold large number of huge size kindle_objs, and plugin's memory usage down to 100~200MB on batch converting.