Add source generator for UnbindAllBindables

[smoogipoo]

Prereqs:

• Refactor source generation for abstraction #5760

Because UnbindAllBindables() is virtual and has been overridden, I've had to make another method InternalUnbindAllBindables() for the SGen.

Before:

Method	Mean	Error	StdDev	Gen 0	Allocated
TestBaseline	0.0000 ns	0.0000 ns	0.0000 ns	-	-
TestCompositeDrawable	2,166.6177 ns	43.0838 ns	44.2439 ns	0.0191	1,640 B
TestContainer	2,571.1976 ns	51.2559 ns	68.4252 ns	0.0191	1,712 B
TestTypeNestedComposite	2,874.6742 ns	57.4976 ns	94.4702 ns	0.0191	1,640 B
TestComplexComposite	3,241.3708 ns	51.8630 ns	53.2595 ns	0.0305	2,672 B

After:

Method	Mean	Error	StdDev	Gen 0	Allocated
TestBaseline	0.0000 ns	0.0000 ns	0.0000 ns	-	-
TestCompositeDrawable	2,069.2690 ns	31.1794 ns	27.6397 ns	0.0191	1,640 B
TestContainer	2,440.0966 ns	12.3803 ns	9.6658 ns	0.0191	1,712 B
TestTypeNestedComposite	2,668.8426 ns	42.0074 ns	39.2937 ns	0.0191	1,640 B
TestComplexComposite	2,589.7359 ns	50.5762 ns	47.3090 ns	0.0305	2,672 B

(note that this doesn't account for reflection memory usage)

[bdach]

Code quality inspections need disabling, probably.

[bdach]

I notice that this seems to be the same thing as in #5761 (comment) again. This means that it is possible for a class to receive two KnownType properties from two disparate sourcegen interfaces with the same value, so I do wonder whether we want to be implementing these case-by-case rather than splitting this one member into a separate interface.

Not sure how much we care about code size but it does look like a potential excess. We may start paying for this in JIT times which we're not going to be able to shave off or hide that easily. On the other side, I also can understand that implementing this to be done as a common thing may not fit in very well into the current syntax target/emitter abstraction.

[bdach]

The Internal qualifier being a prefix rather than a suffix (as we usually do elsewhere) is slightly annoying me here. How intentional was this?

[bdach]

So this means that the generator is going to emit code for every drawable subtype, even if it doesn't introduce any new bindables to unbind? Do we care, and if we do, how much of a pain is it to have it only emit when necessary?

[bdach]

Repeat of #5761 (comment). Should probably be moved out to SyntaxHelpers at this stage.