Performance improvements

[dmtrshat]

Grimoire CSS is already the fastest solution; this work makes it even faster and more efficient, with a safe default and opt-in scaling for multi-core machines.

Overview:

• Improved end-to-end build performance while keeping generated CSS output unchanged.
• Added opt-in parallelism for the build pipeline via GRIMOIRE_CSS_JOBS (defaults to current single-thread behavior).
• Reduced redundant work in the pipeline (fewer repeated passes / less duplicated processing) to improve throughput.
• Lowered allocation/clone pressure in hot paths to reduce CPU time and stabilize memory usage.
• Updated benchmark to track performance deltas and make regressions more visible.

Improvements:

• Wall-time: ~2.68× faster.
• Peak Memory Usage: ~1.37× lower.
• Processing Speed: ~2.68x higher.
• Memory Efficiency: 1.37x higher.

Scaling improves with GRIMOIRE_CSS_JOBS > 1 (best value depends on CPU cores and input size).

---

closes #100