Run CI against macOS and Windows too

[simonw]

Refs:

• LLM 0.13 broken on Windows #407

[simonw]

Got some interesting Windows test failures:

        result = runner.invoke(
            llm.cli.cli, ["chat", "-m", "mock"], input="Hi\nHi two\nquit\n"
        )
>       assert result.exit_code == 0
E       assert 1 == 0
E        +  where 1 = <Result IndexError("Not a valid key: '\\e[d'")>.exit_code

Could that be because Windows wants \r\n?

And a whole bunch relating to paths:

>           assert rows == [
                {"id": "file1.txt", "content": "hello world"},
                {"id": "file2.txt", "content": "goodbye world"},
                {"id": "nested/more/three.txt", "content": "three"},
                {"id": "nested/one.txt", "content": "one"},
                {"id": "nested/two.txt", "content": "two"},
            ]
E           AssertionError: assert [{'content': ...ed\\two.txt'}] == [{'content': ...ted/two.txt'}]
E             At index 2 diff: {'id': 'nested\\more\\three.txt', 'content': 'three'} != {'id': 'nested/more/three.txt', 'content': 'three'}
E             Full diff:
E               [
E                {'content': 'hello world', 'id': 'file1.txt'},
E                {'content': 'goodbye world', 'id': 'file2.txt'},
E             -  {'content': 'three', 'id': 'nested/more/three.txt'},
E             ?                                    ^    ^
E             +  {'content': 'three', 'id': 'nested\\more\\three.txt'},
E             ?                                    ^^    ^^
E             -  {'content': 'one', 'id': 'nested/one.txt'},
E             ?                                  ^
E             +  {'content': 'one', 'id': 'nested\\one.txt'},
E             ?                                  ^^
E             -  {'content': 'two', 'id': 'nested/two.txt'},
E             ?                                  ^
E             +  {'content': 'two', 'id': 'nested\\two.txt'},
E             ?                                  ^^
E               ]

[simonw]

I'd like to fix those test failures, but I also want to ship a 0.13.1 release quickly. So I'm going to mark a bunch of tests as expected to fail on Windows.

[simonw]

Cog check failed:

Line |
   3 |    -p "import sys, os; sys._called_from_test=True; os.environ['LLM_USE …
     |    ~~
     | The term '-p' is not recognized as a name of a cmdlet, function, script file, or executable program. Check the
     | spelling of the name, or if a path was included, verify that the path is correct and try again.

Error: Process completed with exit code 1.