Implement token-saving skill and related evaluation files

[miroslavpojer]

Summary

Adds the token-saving skill — an always-active response-discipline layer that enforces brevity and structured output across all Copilot replies. Also ships the full evaluation suite for the skill.

Changes

Area	Detail
skills/SKILL.md	New token-saving skill definition with trigger rules, line-limit rules, and code-output footer format
skills/token-saving/evals/	Trigger eval + scenario evals covering 5 real-world input files
README.md	Minor clarification
docs/getting-started.md	Minor clarification

Release Notes

• token-saving skill — always-active formatting rules that reduce response verbosity without compromising accuracy. Covers filler suppression, line caps, table/bullet preference, and a mandatory What changed / Why / How to verify footer for code-output responses.

Closes #3

[lsulak]

Review finished. One more thing, a conceptual level below.

The skill says "shortest response that fully answers" but doesn't explicitly say correctness is more important than brevity. An LLM under aggressive conciseness pressure may skip important caveats. Add: "Never sacrifice correctness or safety-critical information for brevity."

[lsulak]

LGTM, good job!