Reduce LSL inlet pull latency

[cboulay]

Summary

• require pylsl 1.18.4b1 and use pull_chunk(..., min_samples=1) to wait for the first sample, then drain immediately available data
• increase the default pull timeout from 5 ms to 100 ms without adding batching latency
• snapshot strong references and scalar pull settings before entering the worker thread
• discard stale results when shutdown or reconfiguration replaces an inlet during an in-flight pull
• update inlet settings documentation and cover all pull paths plus concurrent teardown

The snapshot retains references to the existing inlet, fetch buffer, clock synchronizer, and message template; it does not copy the fetch buffer.

Testing

• uv run pytest -q — 26 passed
• uv run ruff check src/ezmsg/lsl/inlet.py tests/test_inlet.py
• pre-commit Ruff and formatting hooks passed