Q1ASM encoding redundancy

[alecandido]

Since with lark-generated parser, it seems difficult to obtain a satisfactory AST without manually-crafted classes, this generated a bit of redundancy in #868.
Specifically, the Q1ASM instructions are listed twice: in the grammar,

qibolab/src/qibolab/_core/instruments/qblox/q1asm.lark

Lines 15 to 18 in 906da93

	jmp: "jmp" VALUE
	jge: "jge" REGISTER "," IMMEDIATE "," VALUE
	jlt: "jlt" REGISTER "," IMMEDIATE "," VALUE
	loop: "loop" REGISTER "," VALUE

and in the AST classes

qibolab/src/qibolab/_core/instruments/qblox/ast_.py

Lines 52 to 71 in 906da93

	class Jmp(Instr):
	address: Value
	class Jge(Instr):
	a: Register
	b: Immediate
	address: Value
	class Jlt(Instr):
	a: Register
	b: Immediate
	address: Value
	class Loop(Instr):
	a: Register
	address: Value

However, the AST classes are more powerful, since they can even host metadata (e.g. storing documentation), that could be useful in runtime usage (e.g. declared clock cycles).

So, at this point, it seems to be worth to resolve the redundancy the other way round: generate the grammar from the classes.

This could be done by loading the string with placeholders, and replacing the placeholders at runtime from the generated classes.
It is only needed during parser creation. So, it's not going to be limiting even when parsing many scripts (if ever), since it will be done once (it could have been done even at "compile time", if Python had a compilation process - but it's not worth to introduce something like a compilation step just for the sake of this, as it would considerably affect the workflow).