I was kind of curious about the JS payload to the browser, and maybe how the dom node rendering was handled, or if any used canvas/wasm, etc.
I build an expandable Gantt chard for a job in 1999-2000 that was pretty tough going, had to work in NN4.x and IE4/5. It worked, but with NN, you pretty much need to cover the area with a solid ILayer otherwise it'll give someone a seizure with the repaint flickering. Today is so much nicer.
Depending on the level of interactivity, I'd probably try leaning into Rust+WASM with Canvas... As much as I love JS, been running a couple projects playing with Rust WASM and Canvas interactivity and it's been pretty nice. You of course do lose some accessibility though, so that may be an issue depending no your needs.
Most probably, Canvas will not provide a lot of improvement. Existing JS solutions (SVAR, DHTMLX, and Bryntum) already monitor which parts of the final chart are visible and render only those. The algorithm that selects data to render is the real bottleneck on big datasets, not the rendering itself.
As for WASM, the scheduling part may benefit a lot from moving to WASM. From similar projects in the past, it may give a 2x-3x boost in calculation speed, minus the time spent on sync between JS and WASM models. On large datasets, the time for data sending (serialization and deserialization) can consume a significant chunk of the gained boost.