docs(audio): record the episodic half of the $D418 static

[kfox]

docs/architecture/audio.md presented the DAC table/chip mismatch as the whole answer to the audible static on the 4-bit path. It was the steady half.

The episodic half — bursts of 0.5-1.5 s lifting the noise floor 4-10 dB, arriving irregularly, with the floor creeping upward across a long capture — is the Ultimate's Bus Operation Mode sitting at its Quiet firmware default with a TeensyROM+ in the cartridge port. Writes (or Dyn. & Writes) clears it, and nothing under [audio] / [dsp] or in the write path moves it.

The end-user framing already ships in caveats.md and troubleshooting.md, where it reads as a TeensyROM+ link symptom. This adds the architecture side and links across to it, because the two are the same finding seen from different ends: burst statistics gathered under that setting are what made a traffic-coupled mechanism look plausible on the host write path, and so are why the halt-splitting work in the sections above was pursued as an audio-quality fix at all.

With both components accounted for — and neither being bus-halt tick loss — a future measurement that finds episodic noise on this class of rig should check the bus mode before treating the number as being about anything in audio.py.

Also softens the opening sentence of "A DAC table only applies to the chip it was measured on", which claimed the mismatch was the static.

Docs only, no code change. make site-check and tests/test_architecture_index.py pass.