GUACAMOLE-2043: Fix playback of recordings which have multibyte character instructions.

[eugen-keeper]

At first I considered to unite and refactor Guacamole.SessionRecording.loadInstruction and function handleInstruction(opcode, args). My idea was to get rid of Guacamole.SessionRecording.getElementSize. Unfortunately, most likely it will not allow us to fix the issue isolated because the recording can be received as a blob or via a tunnel. loadInstruction/getElementSize are used for both cases and through Gucamole.Parser. There would be too many changes with a chance of new bugs affecting not just the playback.

So to fix the issue isolated, I simply calculate UTF8 string byte size for every instruction content. Performance benchmarks show that a custom implementation of such a function works better than TextEncoder and Blob API. For example, a benchmark. Also, a StackOverflow topic there one of the answers provides benchmarks.

In my tests, a 10Mb recording is preprocessed for about 3 seconds. Byte size calculation of instructions (UTF8 strings) takes about 1-2% of the time. Overall, I did not see significant time difference in preprocessing with or without that byte size calculation. On the other hand, the preprocessing time includes everything including delays to download the content, etc. I could not measure pure parsing time with/without the byte size calculation due to the async function workflow. A single instruction processing takes less than 1 ms. So you cannot measure all instructions separately and then just sum. It will still be 0.

Another option would be to limit byte size calculation for some instructions only. For example,
var getElementSize = function getElementSize(value, unicode) {
var valueLength = value.length;
var protocolSize = (unicode ? getUtf8StringByteSize(value) : valueLength) + 3;
...
and
frameEnd += getElementSize(args[i], opcode === 'name');