Low memory verification batch size (VERIFY_BATCH_SIZE = 5) causes excessive subagent roundtrips and slow runs #814

### Describe the Bug
In `@cortexkit/pi-magic-context` (`src/features/magic-context/memory/verify.ts` / `dist/index.js`), the batch size for memory verification is hardcoded to `5`:

```javascript
var VERIFY_BATCH_SIZE = 5;

// ...
for (let i = 0; i < group.length; i += VERIFY_BATCH_SIZE) {
  batches.push(group.slice(i, i + VERIFY_BATCH_SIZE));
}

### Impact & Evidence
When a project accumulates dozens or hundreds of memories requiring periodic verification:
1. Slicing candidates into batches of only 5 creates an unnecessarily high number of subagent invocations / LLM verification rounds.
2. The verification lifecycle takes disproportionately long (often 2x–3x the necessary runtime), consuming excessive turn latency and subagent coordination overhead.

In practice, modern models with structured output capabilities easily handle verifying 10–20 memories in a single prompt without quality degradation. Raising the batch size (e.g., to 10 or 15) cuts the number of verification rounds and total dreamer verification wall-clock time in half.

### Suggested Fix
Increase the default `VERIFY_BATCH_SIZE` to `10` (or make it configurable via user settings).