Stop to use freed value

Fix wrong free function

mrb_locale_free() should be used for the return value of
mrb_locale_from_utf8().

`mrb_raisef()` only takes `%S` specifier. If you don't have extra
arguments, use `mrb_raise()`.

incorporate fixes from iij/socket

If your system has sa_len but is not BSD-derived, define HAVE_SA_LEN=1
on mrbgem.rake or build_config.rb.

When we have "struct sockaddr_un *s_un", we could not assume *s_un
points to a memory region which size is at least sizeof(*s_un).
Even worse, it may be shorter than sizeof(struct sockaddr) on some
systems.

ref: https://github.com/iij/mruby-socket/issues/45

Ranges are almost immutable but `initialize` and `initialize_copy`
modify the receiver so that we need to add a write barrier.

The problem was caused by `Fiber.current.resume'.

Fix unnecessary rebuild of tests.

This reverts commit c6736357.
The assumption was wrong and caused the issue; fix #4020

Copying all flags from the original object may overwrite the clone's
flags e.g. the embedded flag.

[mruby-socket] Declare inet_ntop and inet_pton as static

fix broken lexical variables

This change is required to support #4038.

This reverts commit b8869498.

To avoid conflicts with multiple definitions of inet_ntop and inet_pton if compiled with # define _WIN32_WINNT _WIN32_WINNT_VISTA

C:/mingw-w64/x86_64-7.2.0-posix-seh-rt_v5-rev1/mingw64/bin/../lib/gcc/x86_64-w64-mingw32/7.2.0/../../../../x86_64-w64-mingw32/lib/../lib/libWs2_32.a(dghfs00169.o):(.text+0x0): multiple definition of `inet_pton'
722C:/projects/iss/mruby/build/host/lib/libmruby.a(socket.o):C:/projects/iss/mruby/mrbgems/mruby-socket/src/socket.c:80: first defined here

define _WIN32_WINNT _WIN32_WINNT_VISTA

fix env->stack misadjusting: fix #4029

`Hash#clone` did not work properly; fix #4030

let Hash#merge keep ifnone value

Since `TT_ICLASS` is a internal object that should never be revealed
to Ruby world.

`exc_debug_info()` and `mrb_keep_backtrace()` raise `FrozenError`
if the exception is frozen and lead to infinite loop.

`then' was added in CRuby 2.6.

Fix test of mruby-pack with big-endian CPUs

When running the mruby-pack test with big endian, test data is incorrect,
so it will fail with "i" and "I".

------
Fail: pack/unpack "i" (mrbgems: mruby-pack)
 - Assertion[1] Failed: Expected to be equal
    Expected: "\xff\xff\xc7\xcf"
      Actual: "\xff\xff\xcf\xc7"
 - Assertion[2] Failed: Expected to be equal
    Expected: [-12345]
      Actual: [-14385]
Fail: pack/unpack "I" (mrbgems: mruby-pack)
 - Assertion[1] Failed: Expected to be equal
    Expected: "\x00\x0090"
      Actual: "\x00\x0009"
 - Assertion[2] Failed: Expected to be equal
    Expected: [12345]
      Actual: [14640]
------

This will fix the test data at big-endian.
Signed-off-by: Nobuhiro Iwamatsu <iwamatsu@nigauri.org>