ARM only allows aligned accesses.  GCC generated unaligned accesses for the copy buffer, but ARM forces alignment, causing the first two bytes to be trampled.  I changed the mechanism to create two temporaries and memcpy them in.  It now works on ARM (and x86) for large websocket chunks.

Update ReleaseNotes.md(spell mistake)

spell mistake

Consistent use of NS_MALLOC, NS_FREE and NS_REALLOC

malloc(), free() and realloc() were used in a number of places.
Added NS_CALLOC as well.

O_BINARY might be defined.

MONGOOSE_NO_CGI needs special handling.

Added size macros.

Added MONGOOSE_NO_USER configuration define.

If socket is about to be closed do not add to [read|write]_set.

If NSG_CLOSE_IMMEDIATELY is on, then the socket will be closed,
which means that (if the socket was added to either read_set or
write_set) the subsequent select will in turn be performed on a
socket that has been closed.

Standard socket implementations seem to ignore a descriptor
referring to a closed socket but LWIP
(http://savannah.nongnu.org/projects/lwip/) crashes. So better
not to add to the sets.

MONGOOSE_NO_CGI is implied by both NS_DISABLE_SOCKETPAIR and
MONGOOSE_NO_FILESYSTEM. Thus, so as not to get complaints about
the same define being defined twice, some extra checks are needed.

On very resource constraind devices you need to be able to
control the size of all buffers and structures placed on the
stack.

Low-end systems, not running a proper OS, may not have the
concept of a user.

Some unix environments defines O_BINARY.

Conditionally include dirent.h and dlfcn.h

If you do not have a filesystem or do not support dav or
dynamic loading, the corresponding header files need not
be included (some environments might not have them).

Added #define's for deprecated POSIX names