Quil-T extension: Named measurements

[antalsz]

This change extends Quil-T with the ability to specify multiple distinct calibrations for measuring a single qubit, allowing us to distinguish MEASURE 0 ro and MEASURE!midcircuit 0 ro. You can read the rendered form of the specification with these changes.

---

Since the discussion of named measurements in #90 has calmed down, it seems like it's time to advance this proposal as a PR which resolves #90.

In particular, I believe the right way to think about named measurements is as a Quil-T extension to Quil, just as we think about all pulse-level functionality. This is indeed the point of having annexes to the main specification; as the specification says in §1.2

There are additional sections, called annexes, which are optional extensions to the Quil language. These extensions may modify the syntax or semantics of the base specification in noted ways.

Thus, I have only modified the main body of the Quil specification by adding a forwards-looking note to §7, "Measurements", indicating that Quil-T modifies these productions. The remaining changes are exclusively to Annex T.

I went with the syntax MEASURE!midcircuit for named measurements, as this seemed to capture the annotation-like form that we wanted and there was some concern about parentheses and brackets being too overloaded to use.

Note that this PR also includes the changes from #96, which can be reviewed separately or simply superseded by this PR.

I'll recapitulate the motivation from #90 below, slightly modified for this new context.

---

Currently, Quil-T only supports one form of measurement: the MEASURE instruction, which measures a single qubit in one of two ways:

⟨Measurement for Effect⟩ ::= MEASURE ⟨Formal Qubit⟩

⟨Measurement for Record⟩ ::= MEASURE ⟨Formal Qubit⟩ ⟨Memory Reference⟩

⟨Measurement Instruction⟩ ::= ⟨Measurement for Effect⟩ | ⟨Measurement for Record⟩

This physical interpretation of this measurement is done by the DEFCAL MEASURE calibration, which can specify one interpretation for each of those two forms of measurement:

⟨Measure Calibration⟩ ::= DEFCAL MEASURE ⟨Formal Qubit⟩ ⟨Identifier⟩? : ⟨Instruction⟩+

For each qubit, then, Quil-T provides only two ways to realize measurement: one way for effect and one way for retrieving the data. However, we somewhat regularly find ourselves needing to support measurements in ways that go beyond this. In particular:

• Midcircuit measurement often needs a lighter touch than end-of-circuit measurement, to avoid destroying the qubit state entirely, so we want a midcircuit measurement of a qubit to be realized with a lighter touch than our end-of-circuit result collection.

• Conditional RESET is implemented in terms of a measurement operation, but one that may benefit from its own physical realization; at a minimum, it wants to look like midcircuit measurement. We do not address how to specify conditional RESET in this proposal, but it would be most natural to build it on this functionality.

In order to solve this, we would like to have multiple distinct physical realizations of MEASURE q for a given qubit q. We indicate these with a !bang; MEASURE!midcircuit 0 ro indicates a distinct measurement operation from MEASURE 0 ro.