Bump default tokenBatchSize from 1 to 4

[corasan]

Summary

• Raise the default tokenBatchSize from 1 to 4 in HybridLLM so streaming emits ~4× fewer TokenEvents, JSON encodes, and JS bridge crossings out of the box, with negligible chunking at typical on-device generation speeds.
• Update all three default sites (property initializer, load() reset, unload() reset) plus the fallback when options.tokenBatchSize is omitted.

Test plan

• Stream a long response in the example app and confirm typewriter UX still feels smooth
• Verify explicit tokenBatchSize: 1 still produces per-token events
• Verify explicit tokenBatchSize: 8 still produces 8-token batches