Write an ADR explaining why stub scores are static rather than derived from on-chain heuristics

[knytcomics-ui]

Problem

A reasonable contributor or downstream consumer might ask: why hand-pick static scores at all, instead of computing them from some heuristic (age of account, transaction velocity, shared-funding-source graph) that could generalize to fixtures automatically? This is a real design decision this repo has implicitly made (per the README: "so scoring can be tested without a live backend") but never explained — without a written rationale, someone will eventually propose "just compute it" as a improvement without understanding why static, hand-labelled ground truth is actually the point.

Requirements

• Add docs/adr/0001-static-stub-scores.md (or similar ADR-format doc) explaining: the goal is a fixed, known-correct ground truth to score against, not a realistic scoring algorithm; determinism and offline execution are required properties (per the README's stated purpose); using a heuristic would make the "ground truth" partially derived from the same kind of logic being tested, undermining its use as an independent check.
• Reference this ADR from the README's Overview section.

Acceptance Criteria

• ADR is written, reviewed, and linked from the README.

Part of the Gryd-lock organization. This repo supplies the labelled testnet fixtures and stub scores consumed by grydlock-oracle-adapter's StubOracle and evaluated end-to-end by grydlock-research.