Add new ideas for IR

[Doomsk]

Previous IR structure and logic were lacking essential working pieces for most of the non-trivial simple instructions. Now it should work to represent all the possible expressions, such as declare, assign, call, call with body, any combination of the previous ones and more.

work in progress.

@q-inho we discuss about this PR later. You can check the code at the new IR file.

I also placed a test so we can check how it looks like and have proper ideas, comments, suggestions, critiques, etc. The test is basically as follows:

hhat_lang/python/tests/dialects/heather/code/simple_ir/test_ir.py

Lines 14 to 42 in 1a23ff9

	def test_dac1() -> None:
	qq = Symbol("@q")
	q1 = CoreLiteral("@1", lit_type="@int")
	qu3 = Symbol("@u3")
	qredim = Symbol("@redim")
	i1 = IRDeclare(var=qq, var_type=qu3)
	assert i1.instr == IRFlag.DECLARE
	assert i1.args == (qq, qu3)
	i2 = IRAssign(var=qq, value=q1)
	assert i2.instr == IRFlag.ASSIGN
	assert i2.args == (qq, q1)
	i3 = IRCall(caller=qredim, args=IRArgs(qq))
	i3_args_block_name = IRBlock(IRArgs(qq)).name
	assert i3.instr == IRFlag.CALL
	assert i3.args == (qredim, i3_args_block_name)
	block1 = IRBlock(i1, i2, i3)
	assert block1.instrs == (i1, i2, i3)
	ir1 = IR()
	ir1.add_new_block(block1)
	assert block1.name in ir1.table
	assert ir1.table[block1.name] == block1
	print()
	print(ir1)

[q-inho]

@Doomsk Thanks for the new IR design you mentioned. It clarifies how to implement the upcoming functions for my side as well. Maybe we could also add test covering CALL_WITH_BODY and CALL_WITH_OPTION to ensure those features work as expected?
Additionally, do you have a further roadmap or todo lists for this new IR?

[q-inho]

I noticed a few spots that appeared in mind while reading the file, so I've left some comment and question.

[q-inho]

Maybe name=IRFlag.CALL_WITH_BODY should be name=IRFlag.CALL_WITH_OPTION?

[q-inho]

Just question. I think Python hash output varies between interpreter sessions, so the same code may produce different block names on each run. Could a stable naming scheme make comparing IR instances and serialization the IR easier?

[Doomsk]

I thought of using uuid.uuid5 at first, but seemed too much for the job. Now that you mentioned it, makes sense to replace to uuid5 indeed

[Doomsk]

Thanks @q-inho ! Indeed, there are things to improve. While I was playing with it, I realized "why not trying to write from source code to IR directly?"
If you are available, we can discuss it further, both the IR roadmap and ways to get it clearer and leaner.