fix(benchmark): fair to reasoning models, fair to router degradation, real sub-cent costs

Three false signals in benchmark results, all seen on a live run:

- Reasoning models were scored ❌ because the per-test max_tokens cap cut
  them mid-<think> before the actual answer. Every run now gets +256 tokens
  of headroom; non-thinking models stop at EOS as before, thinking ones get
  to finish (math-mult on nanbeige: false 0/1 → honest 1/1).
- When the router degraded (timeout → fallback chain) the table showed
  "✗ degraded" and counted it against classification accuracy. There was no
  classification to judge — such rows now show a muted "⚠ router degraded"
  with an explanatory tooltip and are excluded from the accuracy KPI.
- Costs below $0.0001 (cheap flash models) collapsed to "$0.0000". money
  formatting now grows precision until two significant digits survive
  ($0.000035 instead of $0.0000).

2 new strings in all 7 languages.

web/js/i18n/de.js

@@ -616,4 +616,6 @@ export default {
   "keys.h.st.installed": "Auf der Gateway-Maschine installiert",
   "keys.h.st.configured": "Bereits für dieses Gateway konfiguriert",
   "keys.h.st.notFound": "Auf der Gateway-Maschine nicht gefunden",
+  "bm.degraded": "Router degradiert",
+  "bm.routerDegraded": "Der Router hat nicht rechtzeitig geantwortet — die Anfrage nutzte die Fallback-Kette; die Klassifikation wird nicht gewertet.",
 };