Bump rustc to remove ShallowInitBox

[N1ark]

Continuation of #1030
Fixes #973
Related to rust-lang/rust#148190

Added a transformation pass to make vec! safe. For the following original MIR:

((((*_4)).1).0).0 = [move _5]
storage_dead(_5)
_3 = move _4
conditional_drop[{impl Destruct for alloc::boxed::Box<T>[@TraitClause0, @TraitClause1]}<MaybeUninit<[u8; 1 : usize]>[{built_in impl Sized for [u8; 1 : usize]}], Global>[{built_in impl MetaSized for MaybeUninit<[u8; 1 : usize]>[{built_in impl Sized for [u8; 1 : usize]}]}, {built_in impl Sized for Global}]] _4
storage_dead(_4)
vec_2 = box_assume_init_into_vec_unsafe<u8, 1 : usize>[{built_in impl Sized for u8}](move _3)

we want something akin to

_T = [move _5]
_3 = Box::write(move _4, move _T)
_B = unsize<Box<[T; N]>, Box<[T]>>(move _3)
vec_2 = std::slice::{impl}::into_vec(move _B)

[N1ark]

@Nadrieril ready for review :D
fyi i added builtin functions rather than actual function calls for Box::write and Box::into_vec, because by the time the pass runs the TyCtx is lost so we can't retrieve them. Either way I think it makes sense given the other builtin functions (cough cough BoxNew)

[Nadrieril]

I'd prefer to use the actual std functions, is that doable? I'm thinking that when we encounter a call to box_assume_init_into_vec_unsafe during translation we can trigger translation of these two, so that the reconstruction pass can find them later (or even maybe we can directly transform during translation, whichever easiest)

[Nadrieril]

ugh I don't know why our normal path normalization didn't work here, plz just mark this test as //@ no-check-output

[N1ark]

just manually fixed, will do it next time :P

[Nadrieril]

(I really want to get rid of BoxNew, it's an artefact of a time long past ^^)

[N1ark]

@Nadrieril is it ok to leave this as is for now so we can keep bumping the toolchain, and we eventually get rid of all the builtin functions later? ill go through the issues for this PR dw; this is partly because @joshlf is opening a PR to have some sort of automatic toolchain bumper in CI (since often the toolchain bump doesn't cause any breakage)

[Nadrieril]

yes do leave the builtins like this we can make it better later