Improve and uniformize naming across CubeCL (axis/count/line_size)

[adolago]

Refs #353

Summary

This PR applies CubeCL naming standardization across IR, core, std, runtime, and backends.

Implemented from #353

• Use axis instead of dim for tensor-axis naming (Metadata::{Stride, Shape}, tensor/tensormap APIs, std tensor helpers, backend metadata handling).
• Prefer *_count instead of num_* for counts (HardwareProperties, codegen metadata, runtime call sites, local count variables).
• Standardize line_size / lined terminology in place of vectorization wording in public-facing APIs/docs and operation/launcher paths.
• Add tensor ordering aliases:
  • MatrixLayout::IncreasingOrder (alias of ColMajor)
  • MatrixLayout::DecreasingOrder (alias of RowMajor)
• Add conventions.md at project root documenting naming conventions.

Backward compatibility

• Kept existing matrix layout names (RowMajor, ColMajor) and added aliases.
• Kept tensor_vectorization_factor(...) as a deprecated compatibility alias.

Validation

• cargo xtask check format
• cargo xtask check lint
• cargo xtask check audit
• cargo xtask doc build
• cargo xtask doc tests
• cargo xtask test --ci