win: symbol exports and minor fixes

[dhiltgen]

Proposed changes

This PR extracts the non-CUDA portions of #2972 for Windows support.

Test results (known GGUF library incompatibility on Windows)

(.venv) PS C:\Users\danie\code\mlx> .\build\tests\Release\tests.exe 
[doctest] doctest version is "2.4.12"
[doctest] run with "--help" for options
===============================================================================
C:\Users\danie\code\mlx\tests\load_tests.cpp(43):
TEST CASE:  test gguf

C:\Users\danie\code\mlx\tests\load_tests.cpp(43): ERROR: test case THREW exception: [save_gguf] Compile with MLX_BUILD_GGUF=ON to enable GGUF support.

===============================================================================
C:\Users\danie\code\mlx\tests\load_tests.cpp(97):
TEST CASE:  test gguf metadata

C:\Users\danie\code\mlx\tests\load_tests.cpp(97): ERROR: test case THREW exception: [save_gguf] Compile with MLX_BUILD_GGUF=ON to enable GGUF support.

===============================================================================
[doctest] test cases:  221 |  219 passed | 2 failed | 0 skipped
[doctest] assertions: 3045 | 3045 passed | 0 failed |
[doctest] Status: FAILURE!

Checklist

Put an x in the boxes that apply.

• I have read the CONTRIBUTING document
• I have run pre-commit run --all-files to format my code / installed pre-commit prior to committing changes
• I have added tests that prove my fix is effective or that my feature works
• I have updated the necessary documentation (if needed)

[zcbenz]

The documentation generator does not seem to recognize the macro:

/home/runner/work/mlx/mlx/docs/src/cpp/ops.rst:6: WARNING: Error when parsing function declaration.
If the function has no return type:
  Error in declarator or parameters-and-qualifiers
  Invalid C++ declaration: Expecting "(" in parameters-and-qualifiers. [error at 8]
    MLX_API array roll (const array &a, const Shape &shift, const std::vector< int > &axes, StreamOrDevice s={})

[zcbenz]

This does not work on mac:

  /Users/runner/actions-runner/_work/mlx/mlx/mlx/random.cpp:92:4: error: member reference base type '__fp16' is not a structure or union
     92 |   f.bits_ -= 1;
        |   ~^~~~~~
  /Users/runner/actions-runner/_work/mlx/mlx/mlx/random.cpp:128:22: note: in instantiation of function template specialization 'mlx::core::random::below_one<__fp16>' requested here
    128 |         return array(below_one<float16_t>(), float32);
        |                      ^

[dhiltgen]

I couldn't find a good pattern for this one, so I backed it out. As a result, there's one known test failure:

C:\Users\danie\code\mlx\tests\random_tests.cpp(244):
TEST CASE:  test random uniform

C:\Users\danie\code\mlx\tests\random_tests.cpp(368): ERROR: CHECK( abs(float(mean(out).item<bfloat16_t>()) - 0.5f) < 0.02 ) is NOT correct!
  values: CHECK( 0.244141 <  0.02 )

[zcbenz]

I can reproduce this failure on Mac running CPP tests with CPU, and it is also failing on Windows. /cc @awni

[zcbenz]

Properly exporting public APIs on Windows requires adding a MLX_API macro before every public function, which is a common practice. I'm good with that but also /cc @awni @angeloskath @davidkoski since it would change how we define APIs.

[zcbenz]

Looks good to me, thanks for putting this together!

We still need a few more reviews before merging though.

[zcbenz]

I hope there is a better solution but I'm good with what it is now. I have created a issue to track this: #3031.