test(budgets): iron-honest wall-clock budgets — 3x the worst honest-iron measurement

Seven micro-budget tests were calibrated on one fast desktop and failed on
other honest iron with zero functional failures (bisect-proven pre-existing;
David-waived for 10.1/10.2 with this recalibration filed as the cure). Every
budget is now at least 3x the worst measurement observed across three
machines, each with a comment naming its calibration basis; the find-unified
micro-comparison of two sub-millisecond timings becomes a ratio assertion
(absolute equality of microsecond pairs can never be stable). The
inference-bound trim-history correctness test gets a timeout covering its
slowest observed run (174s) — its assertions are exact and untouched.

These remain order-of-magnitude guards; real perf enforcement lives in the
dedicated perf lanes with iron-specific budgets, per the gate-speed standard.

Known non-test artifact, documented not hidden: on slow-inference machines a
minutes-long awaited-embed loop can trip vitest's worker-RPC 60s tolerance
('Timeout calling onTaskUpdate') — all tests pass, vitest exits 1 on the
unhandled orchestration error. The CI lanes on faster iron exit clean; if a
lane ever trips it, the test moves to deterministic embeddings (its
assertions are size-bookkeeping, not embedding quality).

tests/unit/neural/NaturalLanguageProcessor.test.ts

@@ -343,9 +343,11 @@ describe('NaturalLanguageProcessor', () => {
       const duration = Date.now() - startTime
       
       expect(result).toBeDefined()
-      expect(duration).toBeLessThan(200) // Should be fast
+      // order-of-magnitude guard: worst honest-iron measurement 4.8s
+      // (CPU-only inference path, 32-core box); 15s budget covers 3x that
+      expect(duration).toBeLessThan(15000)
     })
-    
+
     it('should handle multiple queries efficiently', async () => {
       const queries = Array(10).fill('Find AI research')
 
@@ -356,8 +358,10 @@ describe('NaturalLanguageProcessor', () => {
       const duration = Date.now() - startTime
 
       expect(results).toHaveLength(10)
-      expect(duration).toBeLessThan(2000) // Should handle batch in reasonable time
-    })
+      // order-of-magnitude guard: worst honest-iron measurement 48.2s for 10
+      // concurrent inference-path queries (CPU-only, 32-core box); ~3x headroom
+      expect(duration).toBeLessThan(150000)
+    }, 200000)
     
     it('should cache pattern matching for performance', async () => {
       const query = 'Find machine learning papers'