Align with official card: 8.3B/1.5B-active, 131K ctx, Pythonic tool-use section

Default max-num-seqs=16, fp8-KV 128K serve example, measured concurrency (~10 full-128K sessions)

Enrich model card: RTX PRO 2000 benchmarks, chat template, concurrency flags, Lna-Lab signature

LFM2.5-8B-A1B NVFP4 (modelopt W4A4) — Lna-Lab