Address monitors by I2C bus and serialize writes per display to cut brightness latency

[knowlen]

Problem

Changing brightness took 1-2 s to reach the monitors, and startup took about 5 s.

• Every ddcutil call addressed the monitor with -d N, which makes ddcutil re-detect all displays first. On a machine with 11 I2C buses that costs 0.9-1.2 s per call.
• The async worker only sent updates every 0.5 s and launched writes fire-and-forget, so writes to both displays contended for the bus (1.7 s for two) and an older write could finish after a newer one, leaving the monitor at a stale value.
• Startup waited 1.0 s between reads and slept 1 s before the UI.

Changes

• detect_displays() records each display's I2C bus; get_vcp_value, set_vcp_value and set_vcp_value_async use --bus N (falling back to -d N if unknown). Writes add --noverify --skip-ddc-checks, reads --skip-ddc-checks.
• AsyncDDCWorker keeps at most one write in flight per display and sends the newest pending value when it finishes; intermediate values are skipped, displays are independent, and writes cannot land out of order. update_interval now defaults to 0.
• _command_interval defaults to 0 (ddcutil already enforces DDC/CI timing); the startup sleep(1) / sleep(0.5) are removed.

Measurements (2 x LG 4K over HDMI, ddcutil 2.2.7)

	Before	After
setvcp per call	0.98-1.21 s (-d 1)	0.10 s (--bus 2 --noverify --skip-ddc-checks)
Both displays in parallel	1.74 s	0.12 s
Slider change -> both monitors updated	1.2-1.7 s + up to 0.5 s batching	0.11 s
Startup (detect + reads)	~5 s	~1.0 s

Tests

• Updated expected ddcutil arguments; added tests for bus addressing after detection, newest-value-wins while a write is in flight, and independent displays.
• 70 passed; black, ruff and mypy clean.

[claude]

Claude encountered an error —— View job

---

I'll analyze this and get back to you.