* Add build-zentf skill for Python/C++/Java builds

Add an agent skill that builds and verifies the zentf plugin from source
across the Python, C++, and Java interfaces. The TensorFlow version is the
mandatory first prompt and drives config selection, validated against
version_configs/tf_2.*. Covers staged build flow with STOP gates, enable+
verify, and benchmark tuning.

* Add AGENTS.md/CLAUDE.md and fix build-zentf skill inconsistencies

- Add root-level AGENTS.md describing the zentf TensorFlow plug-in repo,
  architecture, build system, and the build-zentf agent skill; add CLAUDE.md
  redirecting to AGENTS.md.
- build-zentf SKILL.md: copy the C++ sample into $PKG/examples/ (and create the
  dir) so it matches the compile step that references examples/sample_inference.cpp.
- build-zentf SKILL.md: state gcc requirement as >= 13.1 per the ZenDNN user guide.
- build-zentf SKILL.md: make the TensorFlow-Java checkout TF-version-aware instead
  of implying a fixed TF 2.21.0 commit, noting the ref must match the selected TF
  version's C++ ABI.

* Add a mermaid flow diagram to the build-zentf skill

zentf shares a front-end (confirm the TF version, conda env, pip install
tensorflow, ./configure) and then branches three ways on the interface
before rejoining a common enable-and-verify tail. That shape is easy to
lose in the prose, so add a Flow section near the top with a mermaid
flowchart reproducing the existing flow SVG, including its palette, so the
Python/C++/Java branch and each STOP gate are visible at a glance.

* Move the build-zentf flow diagram into a standalone .mmd file

The flowchart now lives in build-zentf/build-zentf-flow.mmd next to
SKILL.md, matching the convention used by the build-zendnn and
build-vllm-onednn-zendnn skills. SKILL.md links to it and keeps a short
prose summary of the flow.

Signed-off-by: Rahul Jauhari <rahul.jauhari@amd.com>
Change-Id: Iea2038cdf1a5d98d277e0073b568764b4cb744fa