Regressions with non-ASCII characters in path names

It looks like Emscripten has regressed to no longer work if there are difficult characters in the path names - now most recently we have gotten reports over the character [ç](https://en.wikipedia.org/wiki/%C3%87#Computer) on Windows, (Unicode U+00E7, UTF-8 encoding C3 A7).

This character is present as 0xE7 encoding in ISO-8859-1 (Latin 1), ISO-8859-2, *-3, *-14, *-15 and *-16 charsets.

That made me wonder what the current status of testing this kind of behavior is? Back long time ago when Mozilla was maintaining the test suites, the test CIs ran inside a path that contained all kinds of difficult characters (spaces, single quotes, dollar, hash signs, upper 8-bit Latin-1 chars, BMP Unicode chars and non-BMP Unicode chars).

Reading the circleci script, that no longer seems to be the case?

I wonder if this kind of hardening should be redeveloped to the circleci scripts? It would be good to have Emsdk, Emscripten compiler, the .emscripten config and build temp directories each reside inside a difficult path name, so they exercise all the different sources of path name difficulties that may arise.