Add search quality fixtures and ranking benchmarks

[dipeshbabu]

CommandGraph already has lightweight ranking, synonyms, slots, templates, and a 30-command seed set. The next leftover piece should be a small quality harness so search changes do not regress the core apropos experience.

Scope:

• Add fixture queries for the highest-value intents, for example permissions, port conflicts, disk usage, file search, package install, DNS/network checks, git recovery, archives, and containers.
• Assert expected top result or top-N membership where exact ordering should stay flexible.
• Cover both curated command cards and optional man-index merge behavior where practical.
• Keep this deterministic and local-first; no ML or external service dependency.

Acceptance criteria:

• A test or benchmark fixture file defines representative intent queries and expected commands.
• CI can run the checks with the normal test suite.
• Failures make ranking regressions easy to diagnose without requiring manual CLI testing.

Out of scope for this issue:

• Embeddings, RL, cloud models, full learning-to-rank, or broad roadmap work.