v.0.24.0

Added

• Scan to join the rescue hotspot. When the device is running its own hotspot,
  the round LCD gives its whole face to a QR code: point a phone at it and it
  joins, no squinting at a passphrase. The hotspot exists precisely because the
  device is otherwise unreachable, so the panel is the only thing that can hand
  you the credentials. The small mono OLEDs sit this one out - a code that size
  is not something a phone can focus on.
• The certificate now names the address DHCP gave the device. Typing the IP
  used to land on an untrusted page, because only a static address was ever put
  in the certificate. You could click through for the console, but not for the
  video: a browser refuses a wss:// stream to a mismatched certificate and
  offers no way to accept it. The address is recorded when it arrives and named
  from the next restart, exactly as the IPv6 addresses already were.
• The board button now tells you it heard you. Holding it to reset a forgotten
  password used to be a gesture into the void: no light, no message, no way to
  tell whether the button was even wired. The panel now fills a ring around its
  rim as you hold, empties it if you let go early, and says what it cleared when
  it fires.

Fixed

• A WiFi network that disappears no longer locks you out for good. Choosing
  WiFi holds the wired port down, so if the network moved or changed its password
  the device became unreachable by every route at once - and the one setting that
  would fix it lived in the console you could no longer open. The button reset
  always claimed to put you back on Ethernet; it never actually did. Now it does,
  on that very boot.
• Hold the button AFTER the reset, not through it. Every instruction we
  shipped said to hold it down through a power-on. On boards where that button is
  also the chip's download strap - the ESP32-P4 Function EV among them - that
  drops the chip into firmware-download mode instead: nothing boots, and the board
  looks dead. The window is polled seconds into start-up, so holding it through
  the reset never helped anyway.
• The round LCD no longer starves the rest of the device. Its framebuffer
  lives in PSRAM but was not cache-aligned, so the SPI driver could not send it
  where it lay and copied all 115 KB into internal memory for every frame - of
  which this chip has about half a megabyte in total, shared with TLS, USB, the
  network stack, the SD card and the video encoder. The panel logged a failure a
  second, and the H.264 encoder could not get its reference frame: no picture in
  the browser, from a display setting. The buffer is aligned now and goes out in
  bands, so a single frame can never take the device down with it.
• A picture instead of a black rectangle when H.264 cannot start. The
  hardware encoder needs one contiguous multi-megabyte block for its reference
  frame, and after a long uptime PSRAM can be half free and still not have one.
  The encoder is rebuilt whenever the input resolution changes, so that is when
  it bites. It used to retry on every captured frame - thirty times a second,
  forever, logging the same error each time - while the viewer watched nothing.
  Now it says so once, with the actual memory figures - internal and PSRAM,
  free and largest block, because "out of memory" without them sends you
  measuring the wrong heap - and carries on in MJPEG until the next restart.
  Your codec setting is left alone.
• The video stream steps aside during a firmware update. The device serves
  the console, the stream and the upload from one small pool of connections, and
  writes flash with the encoder running. Reported as updates that fail and then
  work on the second or third try (#19 petrn).
• The round LCD no longer runs long text off its edges. A 15-character IP address
  wants more room than a 240-pixel circle has, and the overflow was silently
  clipped rather than shrunk.
• The console no longer offers pins that are already spoken for: the six SDIO
  lines to the WiFi co-processor, its reset line, and the pin that carries SD
  power on the Function EV. Picking one of those used to kill WiFi with no
  explanation - and it was the display driver's own default data/command pin,
  which is why a panel wired there stayed dark.