Cycle buffers

[froody]

Fix for #528

I'm unsure about the correctness of this change but I ran all the unit tests and they all passed. The attached test fails before the addition of _cycle_buffers and passes after adding _cycle_buffers.

My rationale for the fix is that all the sites that were swapping buffers looked like they intended to rotate 3 variables, but actually lose a variable each time:

_src = _dest;
_dest = _tmp;
_tmp = _src;

is equivalent to

uint8_t *scratch = _dest;
_dest = _tmp;
_src = scratch;
_tmp = scratch;

So that any filter which writes to the _tmp buffer will be corrupting its own input.

[FrancescAlted]

Well seen! This produces a warning on the params of the new _cycle_buffers, but I will fix that later on. Thanks @froody !