GH-50774: [CI][Python] Restore bundled archive order for Emscripten

[tadeja]

Rationale for this change

Fix #50774. After #50650, the Pyodide build failed while loading PyArrow with ImportError: dynamic module does not define module export function (PyInit_lib)
The ImportError is misleading because the export is present. The actual failure is an earlier abort during static initialization of libarrow_python.so, in Protobuf descriptor registration.

#50650 changed the order of libraries merged into libarrow_bundled_dependencies.a. Because the archive contains duplicate weak Protobuf/Abseil symbols, the new order causes wasm-ld to select definitions from different archive members and aborts.

What changes are included in this PR?

Restore the pre-#50650 archive order for Emscripten builds before creating libarrow_bundled_dependencies.a.

Are these changes tested?

Yes, test-conda-python-emscripten passes (Local repro, debug and fix verification with ARCH=amd64 archery docker run conda-python-emscripten)

Are there any user-facing changes?

No.

• GitHub Issue: [CI][Python] Pyodide import error #50774

[tadeja]

@github-actions crossbow submit test-conda-python-emscripten

[github-actions]

⚠️ GitHub issue #50774 has been automatically assigned in GitHub to PR creator.

[github-actions]

Revision: f5714c0

Submitted crossbow builds: ursacomputing/crossbow @ actions-50e51eced4

Task	Status
test-conda-python-emscripten

[tadeja]

Although the first crossbow run failed with pre-existing script timeout at the end in the Chrome step, it actually shows problematic part now succeeds:
https://github.com/ursacomputing/crossbow/actions/runs/30844436419/job/91789163035#step:6:6200

> await pyodide.loadPackage('/build/python/dist/pyarrow-26.0.0.dev99-cp312-cp312-pyemscripten_2024_0_wasm32.whl');

Loading pyarrow
Loaded pyarrow

( While in the issue logs Failed to load pyarrow)

The log also shows the export is present in the wheel, meaning that the PyInit_lib ImportError is misleading message.

Locally I traced the cause (with fix reverted failure reproduced), symbolized by patching abort() in pyodide.asm.js to print the stack, relinked with --profiling-funcs

> await pyodide.loadPackage('/build/python/dist/pyarrow-26.0.0a0-cp312-cp312-pyemscripten_2024_0_wasm32.whl');
Loading pyarrow
ABORT:  Error
    at abort (/pyodide/pyodide.asm.js:10:6334)
    at _abort (/pyodide/pyodide.asm.js:10:289408)
    at invoke_v (/pyodide/pyodide.asm.js:10:1224993)
    at wasm://wasm/0267d6e2:wasm-function[13105]:0x531c4b
    at libarrow_python.so.google::protobuf::FileDescriptorProto::SharedDtor(google::protobuf::MessageLite&) (wasm://wasm/libarrow_python.so-0fa2ac3a:wasm-function[48283]:0x2517a8a)
    at invoke_vi (/pyodide/pyodide.asm.js:10:1223725)
    at libarrow_python.so.google::protobuf::FileDescriptorProto::~FileDescriptorProto() (wasm://wasm/libarrow_python.so-0fa2ac3a:wasm-function[48282]:0x2517795)
    at libarrow_python.so.google::protobuf::EncodedDescriptorDatabase::Add(void const*, int) (wasm://wasm/libarrow_python.so-0fa2ac3a:wasm-function[48600]:0x2527cec)
    at invoke_iiii (/pyodide/pyodide.asm.js:10:1223129)
    at libarrow_python.so.google::protobuf::internal::AddDescriptors(google::protobuf::internal::DescriptorTable const*) (wasm://wasm/libarrow_python.so-0fa2ac3a:wasm-function[48861]:0x255572e)
Aborted()
Failed to load pyarrow
The following error occurred while loading pyarrow:
Aborted(). Build with -sASSERTIONS for more info.
[]

The abort happens during static initialization of libarrow_python.so, inside protobuf's descriptor registration.

[tadeja]

@github-actions crossbow submit test-conda-python-emscripten

[github-actions]

Revision: e44aaa2

Submitted crossbow builds: ursacomputing/crossbow @ actions-e7e0e9a62d

Task	Status
test-conda-python-emscripten

[tadeja]

@github-actions crossbow submit test-conda-python-emscripten r-recheck-most

[github-actions]

Revision: a7d0931

Submitted crossbow builds: ursacomputing/crossbow @ actions-c8b74835fc

Task	Status
r-recheck-most
test-conda-python-emscripten

[tadeja]

@github-actions crossbow submit test-conda-python-emscripten

[github-actions]

Revision: 8d456b3

Submitted crossbow builds: ursacomputing/crossbow @ actions-6f85329ef0

Task	Status
test-conda-python-emscripten

[tadeja]

@github-actions crossbow submit test-conda-python-emscripten

[github-actions]

Revision: 18130db

Submitted crossbow builds: ursacomputing/crossbow @ actions-24de2ba338

Task	Status
test-conda-python-emscripten

[kou]

Wow! Thanks!!!

[kou]

Can we remove this condition? (Can we always set CXX_VISIBILITY_PRESET and VISIBILITY_INLINES_HIDDEN?)

[kou]

BTW, can we always use --profiling-funcs for easy to debug? Does it have a large performance penalty? If so, can we enable it only in our CI?