Issue for tracking and coordinating mlx backend work:

mlx.math

• fft
• fft2
• rfft
• irfft
• stft
• istft
• logsumexp mlx - add missing convert_to_tensor  #19578
• qr
• segment_sum mlx - implement segment_sum and segment_max #19652
• segment_max mlx - implement segment_sum and segment_max #19652
• erfinv feat(math): support erfinv on mlx #19628

mlx.numpy

• einsum
• bincount
• nonzero
• cross
• vdot
• nan_to_num
• copy
• roll
• median Implemented median(...) function. #19568 Implement missing functions in mlx backend #19574
• meshgrid Implement missing functions in mlx backend #19574
• conjugate
• arctan2 Added arctan2 operation #19759
• quantile
• imag
• real
• select
• argpartition mlx - add argpartition to numpy #19680
• slogdet
• select
• vectorize
• correlate
• diag mlx - fix diag and diagonal in numpy #19714
• diagonal mlx - fix diag and diagonal in numpy #19714

mlx.image

• rgb_to_grayscale mlx - add rgb_to_grayscale #19609
• resize - mlx - image.resize add crop_to_aspect_ratio argument #19699

mlx.nn

• max_pool
• avg_pool
• conv
• depthwise_conv
• separable_conv
• conv_transpose
• ctc_loss

mlx.rnn

• rnn
• lstm
• gru

mlx.linalg

• cholesky
• det
• eig
• eigh
• inv
• lu_factor
• norm mlx - add linalg.norm #19698
• qr
• solve
• solve_triangular
• svd

mlx.core

• np.ndarray of i64 is being cast to i32 in mlx during conversion if dtype is not passed
• [BUG] mlx crashes with msg - uncaught exception of type std::invalid_argument: [Scatter::eval_gpu] Does not support int64 ml-explore/mlx#1076
• [BUG] np.ndarray of bfloat16 using ml_dtypes is being interpreted as complex64 ml-explore/mlx#1075
• [BUG] arithmetic operations with numpy arrays are not commutative ml-explore/mlx#1066
• [Feature] arctan2 ml-explore/mlx#1065