feat(camera): configurable resolution for vision explain capture

[Copilot]

Vision explain always used the board’s fixed init frame size, so higher-res sensors (e.g. AtomS3R-M12 OV3660) could not send enough detail for text/label reading.

Changes

• Camera::SetFrameSize(name) — optional API; empty string is a no-op; unsupported drivers return false
• Esp32Camera — parse common size names (QVGA/VGA/SVGA/UXGA/…); sensor-only change when FB fits; reinit (with restore) when a larger buffer is needed; keep hmirror/vflip
• MCP self.camera.take_photo — optional resolution argument (default = board setting)
• AtomS3R-CAM/M12 — CAMERA_FRAME_SIZE_NAME in config.h (default "SVGA"); init QVGA then raise OV3660 → board default, GC0308 → VGA
• Docs — mcp-usage (EN/ZH), custom-board, board README
• Host tests — assert API wiring and AtomS3R defaults

Usage

// board config.h
#define CAMERA_FRAME_SIZE_NAME "SVGA"

{
  "name": "self.camera.take_photo",
  "arguments": {
    "question": "Read the label text",
    "resolution": "SVGA"
  }
}

Higher resolutions trade RAM and upload latency for detail; leave resolution unset to keep the board default.

• Fixes [Feature] Configurable resolution for camera vision/explain (currently hardcoded to VGA) #2164

[78]

@copilot Can we obtain the supported resolution modes (VGA, SVGA, etc.) of the camera device in advance, and write them into the MCP description?

[Copilot]

@copilot Can we obtain the supported resolution modes (VGA, SVGA, etc.) of the camera device in advance, and write them into the MCP description?

Done in 38c0a89.

Esp32Camera now reads the sensor max framesize via esp_camera_sensor_get_info() and exposes the supported named modes through Camera::GetSupportedFrameSizeNames() / GetFrameSizeName().

At MCP registration time, self.camera.take_photo builds its description from that list (e.g. GC0308 → up to VGA; OV3660 → up to QXGA) and includes the current default. The optional resolution argument is only advertised when the driver reports selectable sizes.