Bootstrap LLVM build

[tlively]

@juj volunteered to add a bootstrapping step to build a recent LLVM/Clang using the system compiler and then use that clang to build Binaryen. This will unlock using recent C++ features (e.g. coroutines) in Binaryen while retaining the ability to build the full emsdk on older platforms where the system compiler does not support those features.

[sbc100]

Note: I think we only need to do this on macOS

Another note: Hopefully we can just download a prebuilt version rather than building clang twice!

[juj]

I started doing this the other day, adding a simple emsdk manifest field with

diff --git a/emsdk_manifest.json b/emsdk_manifest.json
index ddea07d..ed9512d 100644
--- a/emsdk_manifest.json
+++ b/emsdk_manifest.json
@@ -29,6 +29,21 @@
     "cmake_build_type": "Release"
   },

+  {
+    "id": "llvm-git",
+    "version": "16.0.6",
+    "bitness": 64,
+    "install_path": "llvm/16.0.6",
+    "git_branch": "llvmorg-16.0.6",
+    "url": "https://github.com/llvm/llvm-project.git",
+    "custom_install_script": "build_llvm",
+    "only_supports_wasm": true,
+    "activated_path": "%installation_dir%/%llvm_build_bin_dir%",
+    "activated_cfg": "LLVM_ROOT='%installation_dir%/%llvm_build_bin_dir%'",
+    "activated_env": "LLVM_ROOT=%installation_dir%/%llvm_build_bin_dir%;EMCC_WASM_BACKEND=1",
+    "cmake_build_type": "Release"
+  },
+
   {
     "id": "releases",
     "version": "%releases-tag%",

This worked well, and let me emsdk install llvm-16.0.6-64bit. I was thinking of adding special rules to emsdk.py to look for this LLVM version being installed on outdated systems, and pass that to CMake for building the latest LLVM.. but didn't quite yet find time.

[juj]

Another note: Hopefully we can just download a prebuilt version rather than building clang twice!

At least for my own CI use, I am happy to build clang twice. This is because my CI is stable and not ephemeral, so later runs will see the prebuilt LLVM, i.e. the bootstrap clang will be built only once.

We could look at creating a bundling script from this source-compiled llvm from emsdk as well, and store that as a downloadable.

[sbc100]

I'm hoping you will be the only user, so hopefully no need.