Add CloseRead and closeError test

Closes #98

Both Reader and Writer now will only ever allocate 16 bytes
for their entire usage :)

This is the minimum possible while still preventing misuse
of a EOFed Reader or a closed Writer.

Remove readLoop

Can't go any lower than this afaict. 16 bytes per Writer
and 24 bytes per Reader. go tool pprof agrees with me on
bytes per op but says the allocs per op are 3 instead of 4
and thinks echoLoop is allocating. I don't know.

Lots of cleanup can be performed.

Closes #95

Next is reusing the header structures.

Closes #93

Improve docs

Closes #91

Closes #87

Closes #90

Add WebSocket masking and correctly use Sec-WebSocket-Key in client