Virtual cwd refactoring

[Girgias]

Commits should be reviewed individually.

Motivation

There are two primary motivations:

• Prevent computing strlen() when we already know it
• Use more appropriate types instead of int to convey meaning and intent

To achieve this, refactoring is conducted one step at a time on the various layers that interact with each other:

• the Virtual CWD mechanism
• the stream layer
• the Windows I/O layer

Some other/unrelated minor refactorings are likely going to happen in some commits when I encounter some oddities.

[Girgias]

@cmb69 as don't have a Windows machine and can't test it, is such a function call change even sensible or not?
And if yes, can I also provide a size_t pointer for the 3rd argument so that PHP_WIN32_IOUTIL_CHECK_PATH_W() doesn't need to make a call to wstrlen()?

As seemingly this changes the behaviour of some tests in ways that I don't totally understand.

[cmb69]

I'm not sure whether doing this change in zend_virtual_cwd.c now is a good idea. According to "Make it work, make it right, make it fast", we shouldn't skip the second step. That code is really contrieved, and especially tsrm_realpath_r() is a 500 line long mess. Some "extract functions" refactorings seem to be in order, although that doesn't appear to be simple; both Anatol and I tried, but gave up, a couple of years ago; that doesn't mean that it's not worth to try a fresh attempt.

[Girgias]

I'm not sure whether doing this change in zend_virtual_cwd.c now is a good idea. According to "Make it work, make it right, make it fast", we shouldn't skip the second step. That code is really contrieved, and especially tsrm_realpath_r() is a 500 line long mess. Some "extract functions" refactorings seem to be in order, although that doesn't appear to be simple; both Anatol and I tried, but gave up, a couple of years ago; that doesn't mean that it's not worth to try a fresh attempt.

I'll chip at this when I have motivation, as passing only a char* is far from ideal.