Keep inline for performance critical functions (functions used in loops)

 * The write_str() method does nothing if the length is zero.
 * Replace "struct unicode_writer_t" with "unicode_writer_t"

The length can be read directly from the buffer

"Inline" PyUnicode_Resize(): call directly resize_compact()

accumulator. Directly write characters into the output (don't use a temporary
list): resize and widen the string on demand.

Don't convert uninitialized characters. Patch written by Serhiy Storchaka.

Removed futimens as it is now redundant.
Changed shutil.copystat to use st_atime_ns and st_mtime_ns from os.stat
and ns= parameter to utime--it once again preserves exact metadata on Linux!

Use also directly unicode_empty, instead of PyUnicode_New(0,0).

Remove the fast-path for 1-character string: unicode_fromascii() and
_PyUnicode_FromUCS*() now have their own fast-path for 1-character strings.

Avoid an useless and expensive call to PyUnicode_READ().

Previously, when an error was encountered some processes would not be
stopped until atexit callbacks were run.  On Windows that was too late
to prevent a PermissionError when regrtest tried to remove the temp
directory it ran the tests in.

associated pwd entry.

associated pwd entry.

of the second argument of PyUnicode_New().

 * Create also align_maxchar() function
 * Optimize fix_decimal_and_space_to_ascii(): don't compute the maximum
   character when ch <= 127 (it is ASCII)

 * formatfloat() uses unicode_fromascii() instead of PyUnicode_DecodeASCII()
   to not have to check characters, we know that it is really ASCII
 * Use PyUnicode_FromOrdinal() instead of _PyUnicode_FromUCS4() to format
   a character: if avoids a call to ucs4lib_find_max_char() to compute
   the maximum character (whereas we already know it, it is just the character
   itself)