tls: improve TLSSocket & Server performance

While working on http2 I spent some time digging around tls and I noticed that there were a whole lot of unnecessary closures and function calls throughout. I decided to go through and try to eliminate as many of them as possible. Turns out some of these didn't even need much changed (such as all the _handle events because the _handle already knows which TLSSocket it belongs to).

I tried to limit the churn as much as possible but there definitely is some of it. I've avoided making unnecessary variable adjustments from var to let, const except where the line in question (or related block) was already changing. I also limited all my changes to performance and kept all the logic the same. While the existing tests cover a very solid amount of this module, I didn't want to potentially introduce bugs in uncovered edge cases.

It's a bit hard to get a good measure of performance for these changes as so much of this performance is being limited elsewhere but I created a basic benchmark for TLSSocket creation which clocked in at (and do note that this is limited by the C++ and other code, there's a note re: JS specific numbers further down)

tls/socket.js n=100000     13.48 %        *** 3.554366e-08

The tls-connect benchmark was not much help because it seemed limited by my system's ability to allocate new sockets rather than the actual performance of the code. I couldn't even run a reliable benchmark comparing master to master itself (with 200 runs, I got results anywhere from -10 to 10% with almost no certainty), let alone my code to master.

I also used vegeta to hit a basic TLS server and ran a benchmark for what the max handled requests per second would be at full saturation. With these changes I got a reliable 4% increase across a total of million requests or so.

Also, running the node profiler on the new code indicates that the JS code throughout TLSSocket takes about 50% less time and the JS code within tlsConnectionListener takes about 35% less time (this is across 40 runs of tls-connect.js and 10m session with vegeta). It's a bit hard to run an actual benchmark on it though because so much of it is tied to other code.

I realize this is quite a lot to review so thanks for any and all feedback! Let me know if there's anything I can do to make this more digestible.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• commit message follows commit guidelines

Affected core subsystem(s)

tls