examples: run #9's test plan on hardware

[bdbarnett]

PR #9 (merged as edea99c, 2026-09-07) added nine examples and rewrote uvc_display.py to prefer MJPEG through jpegio. It was merged on a static review: every _usbif call checked against the C module's locals, argument tables and tuple layouts, the usbif.uac_audio / usbif.hid_keyboard / events surfaces checked, all twelve files compiled, and host_enum.py run on desktop CPython (where it reports no capabilities on WSL, as expected). Nothing was run on a board.

The PR's own test plan, still unchecked:

• P4: soundcard.py — Speakers appear on a PC; audio audible
• S3: usb_speaker.py hosting the P4 — tone on the P4 speaker
• hid_keyboard.py / hid_mouse.py against a PC
• hid_host.py with a keyboard (M1 events)
• uvc_display.py with jpegio firmware (MJPEG) and without (YUY2) — this one was a working example before the rewrite, so it is the first to run
• host_enum.py on a board
• usb_serial.py, usb_mic.py, ram_drive.py
• costume_selftest.py still 31/31

Two things a run may surface that the static pass could not: soundcard.py reads I2S pins and codec hooks from board_peripherals under several guessed names and falls back to _output_power / _codec_call; and usb_mic.py / usb_speaker.py read the negotiated format through getattr(..., "format") with a .stream fallback. Whichever branch a real board takes should be the one that stays.