Fix O(n²) performance in ServerEventParser.parse()

[liefran-sim]

Summary

• Fixes quadratic performance in ServerEventParser.parse() that causes severe delays (15-20s) for large SSE events delivered in small chunks
• Reduces complexity from O(n²) to O(n) by using in-place buffer append and tail-region-only separator scanning

Fixes #48

Problem

When receiving a large SSE event (e.g., 1.5MB) via URLSession's ~8KB didReceiveData callbacks:

1. buffer + data creates a new Data allocation every call, copying the entire buffer
2. splitBuffer scans the entire buffer for \n\n on every chunk, even when no separator is present

For ~180 chunks: total work = 8KB + 16KB + 24KB + ... + 1.5MB ≈ 135MB of copying/scanning.

Fix

1. buffer.append(data) — in-place append, amortized O(1)
2. Tail-region scan — only check the newly added bytes (+ small overlap for boundary-crossing separators) for \n\n
3. Early return — skip splitBuffer entirely when no separator is detected (vast majority of calls)

Impact

Metric	Before	After
Per-chunk work	O(buffer_size)	O(chunk_size)
Overall complexity	O(n²)	O(n)
1.5MB event parse time	~15-20s	<100ms

Testing

Verified in production iOS app with SSE flight search responses containing 200-400 inventories (~1.5MB payloads). Events that previously caused 30s timeout failures now parse in under 100ms.

[Recouse]

Looks like a solid change, thank you