GUACAMOLE-2036: Implement the GuacamoleParser within the ReaderGuacamoleReader.

[aleitner]

Implement the GuacamoleParser within the ReaderGuacamoleReader to support multi byte characters when reading instructions.

[aleitner]

We need to benchmark this to make sure there's no degradation of performance

[mike-jumper]

Well, I've thrown together a quick benchmark, and the new version leveraging GuacamoleParser actually seems significantly faster:

[mjumper@mdesktop ~]$ java -cp ./guacamole-common-old-reader.jar:. ReaderBenchmark desktop-vnc.guac 
desktop-vnc.guac ... 18.79 ms / run
[mjumper@mdesktop ~]$ java -cp ./guacamole-common-new-reader.jar:. ReaderBenchmark desktop-vnc.guac 
desktop-vnc.guac ... 8.54 ms / run
[mjumper@mdesktop ~]$

This is just with calls to readInstruction(), which would be expected to have slightly more overhead than read(). The bad news is that things fail when using read() instead:

[mjumper@mdesktop ~]$ java -cp ./guacamole-common-old-reader.jar:. ReaderBenchmark --buffers desktop-vnc.guac 
desktop-vnc.guac ... 13.9 ms / run
[mjumper@mdesktop ~]$ java -cp ./guacamole-common-new-reader.jar:. ReaderBenchmark --buffers desktop-vnc.guac 
desktop-vnc.guac ... FAIL
org.apache.guacamole.GuacamoleException: No instruction available to read.
        at org.apache.guacamole.io.ReaderGuacamoleReader.read(ReaderGuacamoleReader.java:93)
        at ReaderBenchmark.readFile(ReaderBenchmark.java:25)
        at ReaderBenchmark.main(ReaderBenchmark.java:57)
[mjumper@mdesktop ~]$ 

The test file that I'm using is a recording of a VNC session that's about 19M in size. The benchmark just times how long it takes to fully read through the recording, averaged over 100 trials.

Here's the hacky, quick-and-dirty benchmark: https://gist.github.com/mike-jumper/d0939f0c1a4688e9217062788a3ebfba

[mike-jumper]

A few things:

• Good news: it's faster. Bad news: read() seems to fail (see the comment regarding benchmark results)
• This should be against patch.
• The GUACAMOLE-615 issue was part of a past release (1.5.2), so it's now strictly historical. Any new changes will need to be associated with a new JIRA issue.

[aleitner]

A few things:

* **Good news:** it's faster. **Bad news:** `read()` seems to fail (see [the comment regarding benchmark results](https://github.com/apache/guacamole-client/pull/1057#issuecomment-2679632871))

* This should be against `patch`.

* The [GUACAMOLE-615](https://issues.apache.org/jira/browse/GUACAMOLE-615) issue was part of a past release (1.5.2), so it's now strictly historical. Any new changes will need to be associated with a new JIRA issue.

Big relief that it's faster! I just added tests for reader.read and rebased to patch

[mike-jumper]

These changes do fix the issue with read(), though read() is now taking a bit of a performance hit:

I think the performance hit is due to the need to regenerate the protocol representation of the instruction from scratch. The above graph shows:

• Old: The ReaderGuacamoleReader from before this PR.
• New: The ReaderGuacamoleReader from this PR.
• Revised: The ReaderGuacamoleReader from this PR, with additional modifications to avoid regenerating what we've parsed from scratch (a672229).

I'm good with the changes as they stand and think it makes sense to consider further revisions like a672229 in a separate PR.

Unfortunately, it seems impossible to maintain the significant performance improvement of the readInstruction() implementation here without sacrificing the ability to efficiently pass received instructions on to the client ... but we can get close!