fix ml-explore#172 (ml-explore#173)

- steel_attenion.metal (new) was missing from the build

MLXArray (Data, Dtype) constructor , DType size. (ml-explore#158)

* Array init, Dtype codable and tests

Signed-off-by: Valentin Roussellet <[email protected]>

[WIP] adopt new mlx-c api (ml-explore#150)

- adopt new mlx-c api - ml-explore/mlx-c#38
- mlx v0.21.0

v0.18.1 mlx tag (ml-explore#155)

move to v0.18.0 of mlx (ml-explore#137)

* move to v0.18.0 of mlx

- https://github.com/ml-explore/mlx-c v0.0.10
- https://github.com/ml-explore/mlx/compare/v0.16.0... v0.18.0

* turn on additional swift 6 concurrency checks and fix issues
* adopt new mlx_optional_*

Co-authored-by: Awni Hannun <[email protected]>

replacement for asData() to handle strided backing. (ml-explore#130)

* replacement for asData() to handle strided backing.

- fixes ml-explore#117
- asData() now returns a struct with stride information
- takes a policy enum to indicate whether the consumer wants contiguous data

fix to_array behavior with bool -- was not promoting types correctly (m…

…l-explore#127)

* fix to_array behavior with bool -- was not promoting types correctly
* treat Int as Int32 to match python
* add specialized handling for item() to avoid conversions

update to pick up mlx v0.16.0 (ml-explore#115)

- ml-explore/mlx@v0.15.2...v0.16.0
- https://github.com/ml-explore/mlx-c/pull/31/files

V0.15.2 mlx/mlx-c update (ml-explore#101)

v0.15.2 update

- this is mlx-c aligned with v0.14.0 and mlx v0.15.2

- fix x86 release builds -- conditional compile for neon code
- add bitwise ops from previous releases (now present in mlx-c)
- big change is JIT metal shaders -- the build should be faster and smaller

handle non-contiguous backing when reading out MLXArray (ml-explore#96)

* handle non-contiguous backing when reading out MLXArray

- fixes ml-explore#83
- mlx::core::array can have non-contiguous backing
- handle those cases and simplify the readout