kinda; using a keyword instead of library call for suspend is done primarily to constrain suspension to a single stack frame (i.e. the calling continuation is not first class, same as for return), hence it is used to enforce the stackless-ness.

How so? The decision between a stackful and a stackless coroutine is precisely that, a decision. The tradeoff is that storing all registers/stack variables and processor state before a yield and restoring it after a resume is a non-trivial cost. For some applications, that cost may not matter, but for others, it may be critical.

Also, this is less like a goto, and more like a setjmp/longjmp.

aren't most control structures just sugar for goto?