i18n(core): route order-coordinator and token/margin guards through t()

These files bypassed the i18n table entirely, so a user running LANG=en still
got Chinese order logs. Migrating them surfaced structural duplication too: the
five order paths each spelled out their own 'quantity invalid' string, now one
key parameterised by an order-kind label.

Also fixes a type that carried display text as its domain: TrendLabel was
'做多' | '做空' | '无信号', so the engine's snapshot value *was* the Chinese
string and English rendering depended on matching it. Now 'long' | 'short' |
'none', translated at the edge.

Order-coordinator tests asserted the Chinese literals, which is exactly what
made the gap invisible; they now assert the resolved key so they hold in either
language.

39 new translation keys. 271 pass; tsc and oxlint clean.

tests/isolated-margin-guard.test.ts

@@ -1,6 +1,7 @@
 import { describe, expect, it, vi } from "vitest";
 import { IsolatedMarginGuard } from "../src/strategy/common/isolated-margin-guard";
 import type { AccountSnapshot } from "../src/exchanges/types";
+import { t } from "../src/i18n";
 
 const SYMBOL = "BTC-USD";
 
@@ -71,7 +72,7 @@ describe("IsolatedMarginGuard", () => {
     });
     expect(await guard.ensureIsolated()).toBe(true);
     expect(changeMarginMode).toHaveBeenCalledWith({ symbol: SYMBOL, marginMode: "isolated" });
-    expect(logs.some(([, detail]) => detail.includes("已切换为逐仓模式"))).toBe(true);
+    expect(logs.some(([, detail]) => detail === t("log.margin.switched"))).toBe(true);
   });
 
   it("gives up after the confirm attempts run out", async () => {