[SAoC 2023] Replace DRuntime Hooks with Templates Weekly Update #2

[Teodor Dutu]

Hi,

Last week I started working on converting `_d_newarraym{i,}T` to 
templates starting from the `_d_newarray{i,}T` implementations 
from [my previous PR](https://github.com/dlang/dmd/pull/15299/). 
So far I have implemented the new hooks and updated the lowering 
but ran into some test failures in the compiler.

Before fixing them, [this 
bug](https://issues.dlang.org/show_bug.cgi?id=24159) showed up 
and it was caused by my [lackluster lowering to 
`_d_arrayappend{T,cTX}`](https://github.com/dlang/dmd/pull/13495). I worked on these hooks some time ago and replaced the `~=` with the call to the hook in the AST itself. This created issues during CTFE as hooks are generally not interpretable because they often call `libc` functions.

The solution that I'm implementing now is to store the lowering 
of `~=` in a `lowering` field inside `CatAssignExp`. Then CTFE 
can evaluate the original expression and ignore the lowering and 
the glue layer can generate its IR from the lowering and not the 
`CatAssignExp`. The lowered expression of `arr ~= elem` remains 
the same `_d_arrayappendcTX(arr, 1), arr[$ - 1] = elem`.

However, now this is causing a backend error when `arr` is a 
function call, say `foo()`. To avoid calling `foo()` twice, I 
save its return value to a temporary variable and use that 
instead in the `CommaExp` as follows:

```d
foo() ~= elem;

// is lowered to:

_tmp = foo(), _d_arrayappendcTX(_tmp, 1), _tmp[$ - 1] = elem
```

Somehow the backend cannot find the `_tmp` symbol. I am still 
investigating this and how moving the lowering to another 
expression causes it.

Thanks,
Teodor