Issue draft (espressif/esp-vision — "Firmware" template)

Title: [Firmware] image: PNG accepted at load but decode/save fail ("PNG decoder/encoder is not enabled")

Problem summary

image.Image(path) silently accepts PNG files (header metadata such as width/height
is readable, no error at load), but the first operation that needs pixel decoding
fails with RuntimeError: PNG decoder is not enabled, and saving to a .png path
fails with RuntimeError: PNG encoder is not enabled. This contradicts the
documentation, which lists PNG as a supported format (image.Image.to_png in the
image-model concept page) and states that Image.save() infers the format from the
file extension.

The delayed failure is the harmful part: an app can load a PNG asset at boot without
error and only crash later, mid-stream, when the asset is first used. In our
privacy-camera application, a PNG emoji overlay loaded fine at startup and crashed
the running app (after ~34,000 frames) the moment a face was detected and the
overlay had to be drawn. A load-time rejection (like BMP's) would have caught this
immediately.

Environment

• Board model and hardware revision: ESP32-P4X Function EV Board (ESP32-P4)
• Chip, camera, display, and peripherals: ESP32-P4; camera is not involved in this
  repro; no display connected
• Firmware version/commit and download source or CI job: esp-vision catalog
  firmware, commit 3d84b2d (reported by the device firmware check); ESP-IDF
  v5.5.6. Latest catalog release 2026.11.06 not yet tested.
• ESP-IDF version/commit (custom builds): not applicable (catalog firmware)
• Build command and configuration changes (custom builds): not applicable
• Power supply and connection: USB to host (serial + UVC)

Steps to reproduce

1. Flash or boot the firmware above.
2. Upload a PNG image to flash (e.g. /emoji.png, 128x128) and open the REPL.
3. Run the minimal script:

import image
p = image.Image('/emoji.png')            # accepted, no error
p.width()                                 # 128 -> header is read fine
img = image.Image(320, 240, image.RGB565)
img.draw_image(p, 10, 10)                 # RuntimeError: PNG decoder is not enabled
p.save('/tmp_emoji.png')                  # RuntimeError: PNG encoder is not enabled
p.save('/tmp_emoji.jpg', quality=80)      # OK
j = image.Image('/emoji.jpg')             # OK
img.draw_image(j, 10, 10)                 # OK

Reproduction frequency: 100% across REPL and running-application contexts, on cold
boot and after soft reset.

Expected behavior

Either of the following:

• PNG assets load, decode, and save as documented ("format inferred from the
  extension"; to_png listed as a supported output), or
• image.Image(path) rejects a PNG immediately at load with a clear error, like it
  does for unsupported BMP data (OSError: Unsupported format!).

Actual behavior

• PNG loads silently (metadata readable), then decode fails later:
  RuntimeError: PNG decoder is not enabled.
• PNG save fails: RuntimeError: PNG encoder is not enabled.
• JPEG load/decode/save works end to end on the same firmware.

Logs and attachments

• emoji.png — PNG asset that loads but fails to decode (any valid PNG reproduces
  this; the exact asset is available on request)
• emoji.jpg — JPEG version of the same asset that works (control case)
• No crash log: the failure surfaces as a Python RuntimeError in the REPL and in the
  running application.

Workaround and regression

Workaround: convert assets to JPEG before uploading. Unknown whether this worked on
an earlier firmware; first failing version observed is catalog firmware 3d84b2d
(not yet re-tested on 2026.11.06).