optimization: normalize buffer when it becomes empty after data purge

fix bug: frame data read from network was potentially being moved inside buffer (lb_normalize()) before it was processed by caller to read_frame()

add parenthesis to #define'd constants so later multipications work correctly

fix bug with receive buffer handling. Was causing data corruption during communication errors with long frames.

Add KeepAlive support for all TCP sockets.

merge

correct version of --> Allow "" and NULL on TCP host and service (=> IP=INADDR_ANY, and port=0)

Undo changes inserted in previous commit (included 2 unrelated changes)

Allow "" and NULL on TCP host and service (=> IP=INADDR_ANY, and port=0)

merge

change hton() and ntoh() so we don't use (u8 *) to (u16 *) casts. Some compilers incorrectly change the pointer to force alignment with even addresses.

Mario's fix for the pointer alignment problem. See changeset 4: 99009b24d401.

Variables start_addr and count were read from query_packet using function mb_ntoh_safe. It looks like some compilers change the pointer alignment if the first byte starts at an odd address. Because mb_ntoh_safe uses pointers slave address and count (number of registers) were not read correctly from the buffer when several modbus slaves were present in network. In this temporary solution pointer aritmetics is replaced by simple 256 multiplication.

Setting socket option so that TCP Modbus slave can re-use same port immediately after freeing it