Update shell.nix to nixpkgs-24.05 and LLVM 18

[straight-shoota]

Updates nixpkgs in shell.nix to 24.05 release. This release also brings LLVM 18 (and drops LLVM 11) so we update LLVM as well.

Closes #14269, #14592

[straight-shoota]

Like in #14592 there are some spec failures in CI related to iconv.
Somebody should investigate that on a macOS machine.

[HertzDevil]

Apparently Apple replaced the system iconv implementation from an old GNU libiconv to a FreeBSD-based one in 2023, so this is why I recalled not seeing those spec failures on my M2 in the past?

The actual bug is somewhat weird, and I have managed to reproduce it in C as well. String#encode uses an output buffer of 1024 bytes; if a single call to LibC.iconv fails because the output buffer has some space but is insufficient for the next character, then outbuf_ptr and outbytesleft only advance in increments of 32 bytes, even though the correct output bytes are written after outbuf_ptr for the input bytes it can consume. Those extra output bytes are then lost:

# first call consumes 1023 bytes and outputs 992 bytes; all the `y`s are lost
# second call consumes 3 bytes and outputs 2 bytes for the `好`
("x" * 992 + "y" * 31 + "好").encode("EUC-JP").size # => 994

[HertzDevil]

We should pass either -Dwithout_encoding-Dwithout_iconv or -Duse_libiconv. The latter requires pkgs.libiconvReal rather than pkgs.libiconv, but there isn't an associated pkg-config file and the linker flag is only available in $NIX_LDFLAGS?

[crysbot]

This pull request has been mentioned on Crystal Forum. There might be relevant details there:

https://forum.crystal-lang.org/t/cross-compiling-crystal-applications-part-2/7090/12

[HertzDevil]

My bad, it's -Dwithout_iconv rather than -Dwithout_encoding

[straight-shoota]

Now for some reason CI on macOS doesn't even complete anymore. https://github.com/crystal-lang/crystal/actions/runs/11682308922/job/32535751549?pr=14651 is running for 4 hours now.

[HertzDevil]

I think I have seen that outside this PR (https://github.com/crystal-lang/crystal/actions/runs/11856091183/job/33041700399), so I think we're fine?

[straight-shoota]

Without completion of the CI job we cannot be sure the changes in this PR actually work as expected.
We can of course test it locally to give some confidence, but CI is a different environment.

Also the fact that this hangup reproduces consistently makes me think it is related to the change.

[HertzDevil]

I tried applying #15150 and I think CI is progressing now: https://github.com/HertzDevil/crystal/actions/runs/11893760288/job/33139453764

[straight-shoota]

@HertzDevil That seems to have resolved it. So this is ready for review now.

We can probably update to nixpkgs 24.11 soon.

[straight-shoota]

CI on macOS is hanging again: https://github.com/crystal-lang/crystal/actions/runs/11938427884 😞
Looks like we need to revert this.