Flush stdout before calling _setmode on windows

[HampusAdolfsson]

This fixes a partial regression of #773, where line endings in subprocess output are printed as CR CR LF on windows. The problem can be seen with the following build file:

rule MYRULE
  command = cmd /C echo hello

build foo: MYRULE

Using cat from MinGW:

$ ninja.exe | cat -v
[1/1] cmd /C echo hello^M
hello^M^M

Note the double ^Ms.

I've bisected this to #2085, which enabled stdout buffering on windows. When _setmode is called, it seems that any buffered content also has the new mode applied to it (not the mode that was set when that content was written). Therefore, the buffer should be flushed before calling _setmode. This is described under the "Remarks" section here:

https://learn.microsoft.com/en-us/cpp/c-runtime-library/reference/setmode

If you write data to a file stream, explicitly flush the code by using fflush before you use _setmode to change the mode. If you do not flush the code, you might get unexpected behavior.

[digit-google]

The fix looks good to me. It'd be nice if we had a regression test for this as well. Sadly it looks like our misc/output_test.py does not run on Windows, so adding something here would not work :-/ Since we have no easy way to test that, this PR is probably OK.

[Mhunter15]

#2450