Add hook to initialize Julia on-the-fly during thread adoption

[gbaraldi]

The big changes are moving initialization to when someone calls into julia, making it all automatic. (It still supports manually initializing the runtime). I'm not done with this because i still need to figure out how to embed runtime arguments. Currently it's piped as a string but it's not clear yet if that's the best way forward.

We maybe want to make the changes to adopt thread handling only exist behind a flag. Though that potentially doesn't add much because essentially this adds a very predictable null check to ccallable

Edit: This PR has been partially split. It no longer makes any changes to PTLS codegen, and instead just adds jl_pgcstack_default_func to allow an uninitialized runtime to make it into jl_autoinit_and_adopt_thread

The init changes themselves are left to #57498

[topolarity]

Looks like this ironically brokes the trimming test.

Any idea what's up @gbaraldi ?

[topolarity]

Thanks for splitting this for easy review!

I'm not a huge fan of the breaking change to jl_adopt_thread (a public, albeit dangerous-to-use function) - I think it's confusing that it now asks you for a parameter it almost never uses.

Other than that, changes look good to me 👍 thanks @gbaraldi

[vtjnash]

I am not very happy about injecting all of this new code into every cfunction we have. I think you need to actually initialize pgcstack_func_slot with something

[gbaraldi]

Yeah, I was thinking of some way of making this optional on sysimages. I guess we can add an attr to the sysimage module? The pgcstack_func_slot gets initialized by the runtime

[topolarity]

I am not very happy about injecting all of this new code into every cfunction we have.

Is your concern code size, or performance overhead for the extra branch?

[vtjnash]

Size, affecting speed and maintainability if we add other checks and atomics eventually, for what is always the same slow path code

[topolarity]

I took the liberty of pushing a much simpler strategy of just initializing jl_pgcstack_func_slot and using __builtin_return_address for figuring out the library / entrypoint that triggered auto-init.

This no longer requires changes to PTLS codegen