Name the callee zero test cannot execute in test failures

Makes zero test failures name the callee the runner could not execute instead of appending only the test label. Addresses the diagnostic confusion in #428; the underlying limitation (the test interpreters execute only std.testing helpers and pure zerolang functions) is unchanged.

Before:

zero test unknown function: my test

After, for a callee that does not resolve:

zero test unknown function 'myHelper': my test

After, for a real stdlib function the runner does not execute, matching the existing unsupported-construct wording:

zero graph test runner does not implement 'std.mem.span' yet; tests can call std.testing helpers and pure zerolang functions; use live smoke tests for other std calls: std call coverage

Changes:

• Thread the callee name and call node into test_eval_function and pgt_eval_function so the existing null-function guard is the single place the message is produced, in both interpreters. Failures now anchor to the unresolved call's location instead of the test function's.
• std.-prefixed callees get the not-implemented framing; other unresolved names keep unknown function '<name>'.
• Rewrite the skill-data/testing.md paragraph that read this failure as a malformed test label and advised deleting the test; regenerated embedded_skills.inc is included.
• Add a command contract pinning the message and failure.sourceFile for a package test reaching std.mem.*, and adjust two file line budgets in scripts/compiler-metrics.mts.

Verified locally on darwin-arm64: make -C native/zero-c, pnpm run command-contracts:local, pnpm run conformance:local, pnpm run native:test:local, pnpm run native:sanitize, pnpm run workspace:checks all pass. The :sandbox variants were not run locally (they require VERCEL_OIDC_TOKEN); CI covers those paths.