crt: Fix overflow detection in mingw-w64 fgetpos() emulation

Recent Windows versions have different behavior of 32-bit fgetpos and
returns -1 with errno set to EINVAL when position does not fit into 32 bits.

So check for EINVAL and convert it to EOVERFLOW.

This fixes t_fseeki64.c test when compiled against msvcrt20.dll and run on
Windows 10 against the system os version of msvcrt20.dll.

mingw-w64-crt/stdio/fgetpos.c

diff --git a/mingw-w64-crt/stdio/fgetpos.c b/mingw-w64-crt/stdio/fgetpos.c
index aa3feda..8c93dea 100644
--- a/mingw-w64-crt/stdio/fgetpos.c
+++ b/mingw-w64-crt/stdio/fgetpos.c
@@ -26,7 +26,10 @@
    * then fgetpos32() returns -1, sets pos32 to -1 and do not change errno.
    * Non-zero value in [63:32] bits of 64-bit position offset does not trigger
    * any error condition. In other error cases fgetpos32() returns -1,
-   * sets pos32 to -1 and errno to error value. */
+   * sets pos32 to -1 and errno to error value.
+   * Recent Windows versions have different behavior and returns -1 with errno
+   * set to EINVAL when position does not fit into 32 bits.
+   */
 
   errno = 0;
   ret = fgetpos32(_File, &pos32);
@@ -34,6 +37,8 @@
   {
     if (errno == 0) /* This detects overflow when [31] bit is set. */
       errno = EOVERFLOW;
+    else if (errno == EINVAL) /* Value does not fit into 32 bits. */
+      errno = EOVERFLOW;
     *_Pos = pos32;
     return ret;
   }