Consider adding a global `delay`, or otherwise clarifying the spec

[braised-babbage]

The spec makes explicit provisions for global barrier statements, but there is no mention of any equivalent for delay. This could be convenient to add. Strictly speaking, if we supported this in the OpenPulse spec, this would apply to delay instructions within cal blocks, and would not apply to delay instructions outside of cal blocks (which retain the existing OpenQASM semantics). We should poll the TSC to see whether this would be reasonable to support at the OpenQASM level.

[braised-babbage]

Since drafting this issue, I've personally moved away from global barrier statements, when possible. My reasoning here is that we sometimes wish to combine OpenQASM fragments (or subroutines) "in parallel" e.g. if we have some operations on qubits A,B,C and other operations on qubits X,Y, we may want a program on A,B,C,X,Y. Global barrier statements force a relationship between the corresponding gate schedules. Sometimes this is necessary & intended, but in such contexts barrier; is not simply a way to avoid a bit of extra typing -- it's a way to specify programmer intent. I don't see a similar role for delay; (even if symmetry suggests that it could be included).