Cover asymmetric KV cache width pricing

[IvGolovach]

Summary

This is now a narrow regression-test follow-up after current main absorbed and simplified the runtime auto-planning work from the original branch.

• Keeps current main runtime planning behavior, including the conservative unified-KV auto slot ceiling from the recent multi-node fixes.
• Drops the stale RuntimePlanningProfile / public-mesh lane escalation from this branch instead of fighting the newer runtime work.
• Adds coverage that GGUF KV-cache pricing handles asymmetric key/value widths independently.

Why

The maintainer feedback was to update against current main and re-check the shape after the multi-node runtime changes. After rebasing, the right professional scope was not to reintroduce the older runtime planning profile. The remaining useful piece is the model-artifact regression coverage for asymmetric K/V metadata, which protects the capacity math used by the planner without changing runtime behavior.

Diff scope

• crates/model-artifact/src/gguf.rs
  • Adds a regression test for independent K/V cache width pricing.

Compatibility

• Test-only change.
• No runtime behavior change.
• No mesh wire protocol, gossip, protobuf, Skippy ABI, or plugin protocol compatibility change.

Validation

Local validation:

• git fetch --no-tags origin main:refs/remotes/origin/main codex/kv-context-slot-auto-planning:refs/remotes/origin/codex/kv-context-slot-auto-planning - PASS
• git rebase origin/main - PASS
• git diff --check origin/main...HEAD - PASS, no output
• git diff --check - PASS, no output
• git diff --cached --check - PASS, no output
• cargo fmt --all -- --check - PASS
• cargo test -p model-artifact gguf --lib - PASS, 16 passed
• LLAMA_STAGE_BUILD_DIR=/Users/Funtland/Downloads/mesh-llm/.deps/llama-build/build-stage-abi-metal cargo test -p mesh-llm-host-runtime runtime::context_planning --lib - PASS, 9 passed
• cargo check -p model-artifact - PASS
• LLAMA_STAGE_BUILD_DIR=/Users/Funtland/Downloads/mesh-llm/.deps/llama-build/build-stage-abi-metal cargo check -p mesh-llm - PASS
• /opt/homebrew/bin/cargo-clippy clippy -p model-artifact --all-targets -- -D warnings - PASS

Remote validation:

• PR Quality Checks - PASS
• PR Builds - PASS
• PR Docker Build - PASS

Not run:

• Live performance smoke - not required after the final diff no longer changes runtime planning behavior. Current main runtime planning remains covered by targeted runtime::context_planning tests and remote CI.

Rollback

Revert this PR.

Related: #359

[ndizazzo]

@IvGolovach thanks for pushing a fix!

[michaelneale]

@IvGolovach had to make some mega changes to this based on multi node testing I found - might want to update this to main, and check how it performs (I did consult this one but had to simplify it), am still a little confused on how to avoid things overcommitting and falling over, but glad this is looked at