site: output buffering & flushing II

[carloscortonc]

Simplifications of previous version (always writing output to files):

• kernel now has a map of file descriptors
• Each fd has a type (TTY/PIPE) and a buffer
• process.stdout is sent to fd=1, process.stderr is sent to fd=2
• A new FDStack keeps a stack for each kernel's fd. When output needs to be buffered (e.g. a quote containing command-sustitution echo "node-version=$(node -v)", a new fd of type PIPE is pushed into the stack
• After each command, contents of fd=1 & fd=2 are flushed: if the fd's type is TTY, buffer is sent to renderer and then emptied