docs: Rust core spike decision and companion inventories

[arzafran]

What this does

Records the answer to "should we rewrite Programa on gpui-kit for perf and Windows, and should every terminal be a dumb client of a core running elsewhere". Three docs under docs/plans:

1. rust-core-spike.md, read this first. Perf attribution of the Swift shell (idle, hidden output, churn, real keystrokes: all clean, nothing to fix), the gpui-kit + alacritty_terminal spike numbers, the core-plus-clients architecture with personal and org attach modes, and the decision.
2. rust-core-concepts.md, everything a replacement core has to reproduce, with every socket command, CLI command, MCP tool, settings key, and the 10 Ghostty fork patches.
3. t3code-inventory.md, how pingdotgg/t3code splits server and clients, its RPC contract and relay pairing design, and what is worth porting.

Decision in one line

No rewrite. GPUI is viable for a Windows and Linux client. The plan is a headless core (Rust, local or hosted) with draw-only clients, contract first, the macOS app as the first client. Org mode is a product decision.

Test plan

• Docs only, no code. Read rust-core-spike.md "Decision" and confirm it matches the intended direction.
• Spike repo is at ~/Developer/@darkroom/programa-spike on this machine (not in this repo).

[darkestdarky-bot]

Documentation-only PR adds detailed plan and inventory documents. The spike decision record omits the required ## Functional DAG opening for markdown plans/ADRs per cc-settings. Its decision rule section is also missing the explicit criteria list before deferring to a table. Otherwise the documentation is thorough and follows register and structure guidelines.

Review coverage: 500/1150 diff lines supplied. Partial input (truncated: diff, standards). Inline comments are limited to fully visible, valid right-side hunks. Reviewed commit: c36b5683ada13c2fd31d2d5c7d405462783d261c.