WebSocket decoder bug

[SinisterRectus]

Got this error using coro-wrapper and websocket-codec 3.0:

websocket-codec.lua:72: bad argument #1 to 'lshift' (number expected, got nil)

In coro-wrapper:

local item, newIndex = decode(buffer, index)

In websocket-codec:

local function decode(chunk, index)
  local start = index - 1
  local length = #chunk - start
  if length < 2 then return end
  local second = byte(chunk, start + 2)
  local len = band(second, 0x7f)
  local offset
  if len == 126 then
    if #chunk < 4 then return end
    len = bor(
      lshift(byte(chunk, start + 3), 8), -- line 72
      byte(chunk, start + 4))
    offset = 4
...

It seems that byte(chunk, start + 3) is rarely nil.

Will update with more information when I can.

[SinisterRectus]

Seems to be a result of a split chunk, where only a part of one chunk is found at the end of the current buffer. I'm not sure how this happens or why the newer libraries fail to account for it.

[SinisterRectus]

Some debug info while reading a bad buffer:

#chunk	index	start	length	second	len
4096	1	0	4096	126	126
4096	228	227	3869	126	126
4096	435	434	3662	126	126
4096	627	626	3470	126	126
4096	799	798	3298	126	126
4096	1010	1009	3087	126	126
4096	1201	1200	2896	126	126
4096	1458	1457	2639	126	126
4096	1629	1628	2468	126	126
4096	1820	1819	2277	126	126
4096	2077	2076	2020	126	126
4096	2268	2267	1829	126	126
4096	2479	2478	1618	126	126
4096	2677	2676	1420	126	126
4096	2848	2847	1249	126	126
4096	2984	2983	1113	126	126
4096	3183	3182	914	126	126
4096	3354	3353	743	126	126
4096	3527	3526	570	126	126
4096	3716	3715	381	126	126
4096	3886	3885	211	126	126
4096	4095	4094	2	126	126

This is definitely a result of there not being enough data in the buffer to read. I noticed in 27a3d962e1fa271df40271c5d4fe1dc9bc62c8c9 that you forgot to account for the still-existing, already-read part of the buffer when checking the length of the chunk to be read. I'm assuming that this is the cause of this issue, so I'll go ahead and open a PR for this.