feat: redesign GPU capacity dashboard

feat: HF OAuth sign-in, probe endpoints restricted to Hugging Face org members

feat: add g7e family (RTX PRO 6000 Blackwell)

fix: reload open panel after any probe of the displayed type

fix: move history disk writes to a dedicated writer thread (FUSE append froze the sweep loop)

feat: probe history persisted as daily JSONL + side panel with availability chart

feat: add p6-b300 family and show GPU model under instance type

feat: persist state in mounted HF bucket and skip startup sweep when state is fresh

fix: N.O label instead of N.A for not-offered regions

feat: show N.A label for regions where a type is not offered

feat: AWS GPU capacity checker with ODCR probing