fix: read sink state once per tick instead of three pactl calls per dial

[NyleGarcia]

Problem

With dials on screen, the plugin spawned ~29 pactl processes per second continuously, loading pipewire-pulse.

Cause

_tick_levels runs at ~8 Hz and, for every encoder, _read called sink_exists (pactl list short sinks), get_volume and get_mute — three subprocesses per dial per tick.

Scope

• owdeck/graph.py: replace sink_exists/get_volume with sinks() — one pactl -f json list sinks returning {name: (volume, muted)}; failure or bad output returns {} (mix shows unavailable, as before).
• plugin.py: _sink_states() caches that map for SNAPSHOT_SECONDS (0.9 s), shared by all dials; invalidated after our own set_volume/toggle_mute and on OpenWave pushes, so user turns stay immediate and external changes show within ~1 s.
• tests/test_plugin.py: fixture stubs graph.sinks; new TestSinkPolling bounds pactl spawns across many ticks with multiple dials.

Verification

• make test: 121 tests OK; the bound test fails (108 spawns vs limit 4) with the cache disabled. make validate clean.
• Live: 4 mix dials for 5 s → 6 pactl spawns (~1.2/s, was ~29/s); dials show correct volume/mute.

[github-actions]

🎉 This PR is included in version 0.7.2 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀