gpt-4-turbo model ID, closes simonw#493

Windows readline fix, plus run CI against macOS and Windows

* Run CI on Windows and macOS as well as Ubuntu, refs simonw#407
* Use pyreadline3 on win32
* Back to fail-fast since we have a bigger matrix now
* Mark some tests as xfail on windows

Release 0.13

Refs simonw#325, simonw#351, simonw#373, simonw#376, simonw#394, simonw#396, simonw#397, simonw#404

Closes simonw#402

Release 0.12

Refs simonw#323, simonw#324

Release 0.11.2

Refs simonw#327

Release notes for 0.11.1

Refs simonw#317, simonw/llm-gpt4all#16

Release 0.11

Refs simonw#273, simonw#274, simonw#275, simonw#280, simonw#284, simonw#286

Release 0.10

Refs simonw#225, simonw#229, simonw#231, simonw#254, simonw#256, simonw#259

Release 0.10a1

Refs simonw#229, simonw#244, simonw#247, simonw#248, simonw#254, simonw#256, simonw#259, simonw#263

Don't update stable docs on preview releases