Add type-instability warning to getproperty docs

[drbassett]

I was recently tripped up by a performance problem with a custom implementation of getproperty. In summary, the problem is that using the dot operator within getproperty invokes getproperty recursively, which complicates the job of the type inference system in Julia, and causes type-instability. See the issue I posted to discourse for more details: https://discourse.julialang.org/t/type-inference-problem-with-getproperty/54585.

Using the dot operator is the obvious way to implement getproperty, especially considering the example in the official documentation does so. I think many other users will encounter this same problem, and many of them will be unaware of it. I am guessing it will be difficult to improve the optimizer to handle recursive getproperty calls more intelligently, so instead I propose adding a documentation blurb to warn users about this problem, and how to work around it, which this PR does.

[timholy]

The only reservation I have about merging this is whether the inference problem can instead be fixed. Has it been reported separately?

If this is hard to fix, I'd suggest we also change the example in the docstring to use getfield everywhere.

[drbassett]

Agreed that it would be better to fix the problem than to just document it. As far as I can tell, there does not seem to be an existing issue for this problem: https://github.com/JuliaLang/julia/issues?q=is%3Aissue+is%3Aopen+getproperty+in%3Atitle. I will create one to see if I can get a discussion going.

EDIT Issue is here: #39563

[nalimilan]

Suggested change

	example, replace `a.b` with `getfield(a, :b)`.
	example, replace `a.b` with `getfield(a, :b)` in the definition of `getproperty` for your type.

[vtjnash]

It could be fixed. Is it common enough though to warrant changing? The example already shows how you would typically call getfield in your definition, since the expectation is that if you are overloading getproperty, it is because you do not want it to just forward to getfield, so you need to use the "inner" API anyways.

[JeffBezanson]

I'm not a big fan of enshrining "folk wisdom" about how the compiler works in documentation --- things can easily change and we'll never remember to update it. I think it's enough just to recommend getfield in general for implementing getproperty.

[vtjnash]

I think we now allow some recursion here (though infinite recursion may always be tricky). The issues posted in that discourse seem to be gone anyways.