On STM32, in case of PHY failure, usually because of the lack of 25/50 MHz REFCLK input, the Ethernet block would lock up in a while loop waiting for readiness, which will never happened.  This prevented the board from starting at all in the case of a PHY failure.  This commit adds a retry loop to avoid bricking the board if the Ethernet block is unable to start.

Fix some bugs in networking

* netdb: Fix boundary conditions in dns_recv_response()

Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>

* Fix typos in include/nuttx/net/dns.h

Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>

Approved-by: Gregory Nutt <gnutt@nuttx.org>

Per Matias Edlund:  I recently used the tcflush function and found that it failed with ENOTTY. The reason is that the TCFLSH case in the function uart_ioctl in the file drivers/serial/serial.c does not update the ret value."  On success, the TCFLSH IOCTL logic needs to explicitly set the return value (ret) to zero.

tools/gencromfs.c:  Was duplicating all mode bits in source directory in CROMFS directory.  Some make no sense.  In particular, all write-able bits are cleard unconditionally.  Executable bits are still copying because there is really no way for the tool to know if the file is executable or not.

fs/cromfs:  Block length only needs to be uint16_t, not uint32_t.  Add pading to node structure to assue that alignment is the same on all platforms.  tools/gencromfs.c:  Fix the target offset of the '.' hard link.

fs/cromfs:  Fix a error in reading partial compressed blocks.  The LZF decompressor does not support that operation.  Instead we have to decompress full block into a temporary buffer and copy out the parts that we need.  To compensate for the performance hit, a caching mechanism was added so that we do not have to read the same block repeatedly.  Unrelated:  Also updates some README files.

configs/cc3200-launchpad:  I don't think that there is any reason for BINFS to be enabled in the NSH configuration.

STM32, STM32L4: add one bit read and write to 1-wire interface.  This is needed for devices that require single bit timeslot generation and for 1-Wire search algorithm triplets.

Remove configurations using canutils/libuavcan and examples/uavcan.  libuavcan has not built for a year or so.  The basic problem is that as NuttX advances, the old frozen versino of libuavcan has become absolute because it violates the portable POSIX OS interface.  No one is maintaining the port so there is no alternative but to remove it.

Documentation/NuttxPortingGuide.html:  Some updated wording of description of the directories need for a complete board configuration.

i2c/i2c_master.h:   Change incorrect comment about I2C_TRANSFER return value to indicate that the method returns zero on success rather than the number of bytes transferred.

drivers/sensors/sht21:  Change I2C_TRANSFER return value so that it return on zero on success, not the count of bytes transferred.

Redo commit 7244ae9a.  The same trivial change was done with an online edit, but this resulted in some corruption of special characters in the file, likely because of differences with the character set used by the Bitbucket online editor.

This reverts commit 7244ae9a.

tools/genromfs.c:  Improve check for empty directories.  If the directory contains only unsupported, special files, then it is still empty.  Fix some collection of statistics that was cause statfs() to return some incorrect values.  Add logic to handle the case where the host and the target differ in endian-ness.  fs/procfs:  In show mountpoints, add support so that the CROMFS file system type is recognized.