Also, It would be nice to have buttons and hotkeys in Tesla Term also, that way you don't have to go through the drop down menu to do stuff. (the tesla coil messes with my track pad)

This is something I have considered previously (mostly for the "kill" button), I just never got around to implementing it. I am currently in the middle of a larger refactor in TT (connecting to multiple coils with a single TT instance), once that is done I can look into hotkeys. Can you open an issue for it at https://github.com/malte0811/Teslaterm so I do not forget about it? In general hotkeys are going to be a "fallback" option though, with a mouse being the intended method of interaction. Have you tried using a "traditional" mouse rather than a touchpad? In my experience this is much more reliable around coils.

Since I'm already posting here: Currently Teslaterm has the option to run as a "headless" server (e.g. on an RPi installed in the coil), which the actual control computer then connects to using a standard browser. Is anyone actively using this mode ("node")? It complicates the code significantly in some places, so I would like to remove it if possible.

TOSLINK works quite well already (without Manchester encoding), but some receivers work better than others. I use 2 cheap receivers I got from Netzpfuscher. They work reliably at 500k baud, but when I tested them at ~115k, the signal at the receiver went high too soon when transmitting a character with a long chain of 0 bits. I assume that with less suitable modules this happens quicker, so it's a problem even at 500k baud.
Manchester encoding could fix this, and since the modules (at least the ones I have) seem to be fine with a constant "high" signal it would be possible to use a standard UART module at the PC and do the Manchester encoding in software (effectively using 4 bits per byte).

For the "real" Manchester encoding Netzpfuscher implemented I believe the plan is to use another PSoC at the PC end.

Steve Conner tried this (or something close to it) in 2011, without success (http://4hv.org/e107_plugins/forum/forum_viewtopic.php?128784.0). The thread quickly gets away from the topic and moves to experimental tuning modes (the zero in the transfer function and even more unusual things). As I understand it the bad performance is probably due to low ramp quality. Another possibility (loneoceans mentions this in his QCW1 page) is that the primary current envelope doesn't need to be linear. He measured a roughly quadratic envelope with his QCW2 (might be 1.5?), though I recall seeing someone else (Steve Ward?) showing measurements with nearly constant primary current.
Something I'd be interested in seeing would be a 'traditional' buck-modulated QCW, but taking the buck feedback from the primary current rather than the bus voltage. There are probably good reasons why this hasn't been done, I don't have any experience with or proper knowledge about buck converters.