[libc] Fix the unamed pipe close issue

Signed-off-by: armink <armink.ztl@gmail.com>

There is a memory leak issue when close a pipe.

1. Add macro check in rtdebug.h;
2. Use RT_USING_POSIX for poll/select, stdin etc.
3. Split dfs_posix.h to dfs_posix.h, dfs_poll.h and dfs_select.h;

1. Change 'rt_uint8_t' type of cmd to 'int';
2. Add waitqueue;
3. Split device ipc header files;

Remove dataqueue/pipe if not enable RT_USING_HEAP, supply the missing `not`.

fix: completion wait after, maybe can't clean the flag.

OXape reports: http://www.rt-thread.org/phpBB3/topic3939.html
          driver/ringbuffer: when data length is bigger than space length
          for rb, the read_index in rb is not update.
    When length > space, update the read_index, the bug can be fix, and
          works fine.

When forwarding rx/tx callback from the underlaying device(pipe), the
"dev" argument should be the portal. So the portal callback could think
itself as called.

It does not make sense to open portal without a flag. One should open a
portal with RT_DEVICE_OFLAG_RDWR in most cases.

Portal is a device that connect devices. Currently, you can only connect
pipes in portal. Pipes are unidirectional. But with portal, you can
construct a bidirectional device with two pipes.

Pipe have many differences with char device. The main difference is
altough pipe have both read and write, it is unidirectional in nature.

Some reader(FinSh again) would like to block on a semaphore which is
released in rx_indicate. So we should invoke rx_indicate in
_rt_pipe_resume_reader.

The previous implementation will always blocks the reader/writer.
However, at least FinSh would expect the device to be nonblocking ---
read should return 0 when there is no data in it.