Get thread locale in pj strerror

[LeonidGoltsblat]

This changes are for users who use pjsip under non english Windows redactions.

1. PJ_ERR_MSG_SIZE
   For example russian language is not so laconic as English. Windows FormatMessage function often requires more than 80 character for message buffer. Thats why I need to increase PJ_ERR_MSG_SIZE value (in fact I use 160 - twice more!), but I preffere to declare PJ_ERR_MSG_SIZE in my config_site.h file. The only change I need for this purpose is to surrond #define PJ_ERR_MSG_SIZE by #ifndef

2. PJ_STRERROR_USE_WIN_GET_THREAD_LOCALE
   pj_strerror() calls Windows FormatMessage function with the value of languageId parameter meaning "User default language" which is language configured for user session. This is not always comfortable. For example pjlib_project errno_test() failes for non English user language because FormatMessage return national language string (for examle cyrillic string) not containing the word "invalid" which is test criteria.
   But starting from Windows XP application can set locale (and language) information on application and thread basis. It's logically to expect the program will reflect to current thread locale. My idea is to use current thread language to obtain OS error description.
   To support this I have modified platform_strerror() (for Windows platform) so that it calls FormatMessage with current thread language instead of user default language.
   For compatibility reason I have introduced PJ_STRERROR_USE_WIN_GET_THREAD_LOCALE macro wich may be set in config_site.h file. By default behaviour is not changed, but if user define PJ_STRERROR_USE_WIN_GET_THREAD_LOCALE platform_strerror() will call GetThreadLocale() to get current locale (and language).
   As an usage example: if that macro is defined, error_test() fors English thread locale before calling pj_strerror() and now test is succeeded under non English Windows redactions.

[CLAassistant]

All committers have signed the CLA.