Revive the HelloiOS app

[abadams]

This resurrects the HelloiOS app. Alex and I would really appreciate someone with xcode and an iphone or ipad device taking it for a spin on a real device, because we had a hell of a time just getting it right for the two of us. Start by building and installing (to a local folder) Halide via cmake and then check the README in the app.

Still to do before merging is making CI build the app for simulator on mac.

Fixes #8730

[alexreinking]

We've noticed a problem here: it seem the Xcode dependency analysis is missing a link somewhere and it requires two builds for the generator output to update after making a change.

[alexreinking]

Approving with the caveat that this isn't perfect, see #9049

[abadams]

Haven't merged because this still isn't tested on CI. I don't see how to integrate this into the cmake apps testing setup and it's not really testable via the makefile (because it relies on a cmake install package of Halide). I could have "make distrib" produce a cmake-compatible install package, but that's pretty perverse. I also don't want to make a bespoke runner just for this one app.

[alexreinking]

Haven't merged because this still isn't tested on CI. I don't see how to integrate this into the cmake apps testing setup and it's not really testable via the makefile (because it relies on a cmake install package of Halide). I could have "make distrib" produce a cmake-compatible install package, but that's pretty perverse. I also don't want to make a bespoke runner just for this one app.

I added a workflow to test the app here.

[alexreinking]

fuzz_bounds failures:

Seed: 8412040097261433877
can't prove upper bound: (1 <= 0)
a = (uint1)1
b = (uint1)1
c = (uint1)0
d = (uint1)0
e = (uint1)1
int32x2(uint1x2(ramp(int32((uint1)c), int32(uint1(int32((uint1)e)) && (uint1)0), 2)) || uint1x2(ramp(int32((uint1)e), -25, 2) % int32x2(ramp(uint32((uint1)b), (uint32)4294967286, 2))))
[0, 0]
In vector lane 1:
int32(uint1(int32((uint1)c)) || uint1((int32((uint1)e) + -25) % int32(uint32((uint1)b) + (uint32)4294967286))) -> 1
scope {
	a : [(uint1)1, (uint1)1]
	b : [(uint1)1, (uint1)1]
	c : [(uint1)0, (uint1)0]
	d : [(uint1)0, (uint1)1]
	e : [(uint1)0, (uint1)1]
}

and

Seed: 412326383347904392
can't prove upper bound: ((uint16)1 <= (uint16)0)
a = (uint8)61
b = (uint8)79
c = (uint8)36
d = (uint8)94
e = (uint8)104
uint16x4(int32x4(ramp(x2(uint32((uint8)e)), x2((uint32)4294967174), 2)) > x4(int32((uint8)a)))
[(uint16)1, (uint16)0]
In vector lane 0:
uint16(int32(uint32((uint8)e)) > int32((uint8)a)) -> (uint16)1
scope {
	a : [(uint8)61, (uint8)61]
	b : [(uint8)77, (uint8)98]
	c : [(uint8)7, (uint8)37]
	d : [(uint8)88, (uint8)95]
	e : [(uint8)82, (uint8)113]
}