[Bugfix] Use TCP liveness probe for LMCache server

[EltonChang1]

Fixes #1057

What changed

• replace the cache server exec liveness probe with a Kubernetes TCP socket probe on cacheserverSpec.containerPort
• update the focused Helm assertion for the rendered probe
• correct the cache server probe documentation and defaults

Why

The current LMCache image does not contain the referenced health_probe.py script. The cache server exposes a TCP protocol, not an HTTP health endpoint, so probing its listening socket avoids restart loops caused by an image-specific script path.

Validation

• helm lint helm
• focused cache server Helm unit suite: 2 tests passed
• rendered the cache server deployment and verified tcpSocket.port: 8000
• repository YAML, EOF, trailing-whitespace, and codespell hooks on touched files
• markdownlint-cli 0.44.0 on helm/README.md
• git diff --check and git show --check HEAD

Limitations

The full Helm unit suite reported 142 passing tests and two unrelated failures in the existing chat-template and RayCluster suites under the local Helm/plugin versions. No live Kubernetes or LMCache image test was run.

• Relevant pre-commit checks pass
• Commit is signed off for DCO
• PR title is classified

This draft was prepared with AI assistance and has not yet been human-reviewed.

[gemini-code-assist]

Code Review

This pull request simplifies the cache server's liveness probe by replacing the Python-based exec probe with a TCP socket probe on the container port. These changes are applied to the Helm deployment template, updated in the test suite, and documented in the README. There are no review comments, and I have no additional feedback to provide.