Copybara import of the project:

[copybara-service]

Copybara import of the project:

--
041763c by tinhien11 thanhtinpk092007@gmail.com:

subgraph: validate filter shape dims[1..3] in conv/deconv/depthwise-conv

xnn_define_convolution_2d validated only filter_value->shape.dim[0]
(== group_output_channels * groups) but not dims[1..3] against
kernel_height, kernel_width, and group_input_channels. When a filter
tensor is declared with dims[1..3] smaller than the convolution's
kernel parameters, weight packing (xnn_pack_f32_conv_goki_w) reads
nc * ks * kc elements from the filter buffer — where ks comes from
the convolution parameters, not the filter's declared shape — causing
a heap out-of-bounds read during xnn_create_runtime_v2.

xnn_define_deconvolution_2d and xnn_define_depthwise_convolution_2d
had no filter shape validation at all.

Add full 4D filter shape validation to all three operators, rejecting
mismatched shapes at define-time with xnn_status_invalid_parameter.

Fixes #11118

--
c2ec208 by tinhien11 thanhtinpk092007@gmail.com:

subgraph tests: fix filter shapes rejected by define-time validation

The new conv/deconv/depthwise filter shape checks reject the dummy
geometry used by three structural tests that never execute the graph:

• SUBGRAPH_FP16.convolution_{weights,bias}_used_by_another_node declared
  a 1x1 filter ({2, 1, 1, 3}) against Kernel{3, 3}; use a 1x1 kernel so
  the shapes are self-consistent. Output extents are unchanged.
• SUBGRAPH_NCHW.bottleneck declared filter Remove redundant #includes #10 as {4, 1, 1, 4} for a
  convolution with group_input_channels=8; use {4, 1, 1, 8}.

FUTURE_COPYBARA_INTEGRATE_REVIEW=#11119 from tinhien11:fix/filter-shape-validation c2ec208