Factory view: sortable columns, %-met severity, collapsible sections (late-game scale)

[ApocDev]

The Factory ledger reads fine at ~45 goods but degrades structurally at late-game Py scale (hundreds of items). Rework the goods sections on TanStack Table (headless — markup, badges, drawer-clicks, and mobile card-stacking stay exactly as they are; only the data model moves).

Problems today

• The deficit sort conflates magnitude with urgency. Everything sorts by absolute net/s, so bulk fluids (-500 Water) permanently own the top while a fully-starved 0.3/s intermediate that is killing a science chain sits buried. The severity axis that matters is relative: fraction of demand met.
• No sorting controls at all — and the fixed order is backwards for Surpluses (inherits one global ascending-net sort, so the smallest surpluses list first).
• Nothing collapses. Four always-open sections plus Machines means scrolling through everything to reach anything.
• actual/s — the "factory is actually dying" signal — is the least prominent column.

Plan

1. Add @tanstack/react-table; shared column defs for the goods sections (name, produced/s, consumed/s, net/s, %-met, actual/s), getSortedRowModel, controlled sorting state.
2. Click-to-sort headers with per-section defaults: Deficits by %-met ascending (fully starved first), Surpluses by net descending, Stock buffers by name. Sorting choices persisted (localStorage).
3. %-met on deficit rows (produced ÷ consumed, small percent or thin bar) so 0%-starved reads as alarming regardless of magnitude.
4. Collapsible sections with counts in the header, remembered state.
5. Machines card migrates to the same engine if low-friction, else follows later.

Deliberately deferred: kind/category filter chips and grouping (real but heavier), and @tanstack/react-virtual — the designated escape hatch if row count ever measurably hurts rendering; nothing in this design blocks bolting it on (uniform-height rows).

Bundle size is a non-concern (desktop app; the icon atlas dwarfs the JS).