feat(esp_jpeg): Added option to get image size without decoding it

[tore-espressif]

Change description

Added new public API to get image size without decoding it.

This is very useful for scenarios when we have a JPEG encoded image, but we do not know what resolution does it have, thus we don't know how big output buffer we need for decoding. This scenario is added to test-cases.

This is the last missing feature, before we can integrate this component to esp32-camera, in espressif/esp32-camera#740

[tore-espressif]

cc @igrr since you reviewed some esp_jpeg PRs in the past :)

[github-actions]

Test Results

15 files  15 suites   1m 3s ⏱️
 5 tests  5 ✅ 0 💤 0 ❌
25 runs  25 ✅ 0 💤 0 ❌

Results for commit 3f723c5.

♻️ This comment has been updated with latest results.

[igrr]

May I suggest extending esp_jpeg_image_output_t with a buffer_size field, and instead of esp_jpeg_get_output_size adding something like esp_jpeg_get_image_info, returning esp_jpeg_image_output_t? Then we can get both the buffer size and image dimensions before actually calling the decode function.

Similar to what libpng allows doing:

idf-extra-components/libpng/test_apps/main/test_libpng.c

Lines 30 to 38 in c37fc0a

	TEST_ASSERT(png_image_begin_read_from_memory(&image, buf, buf_len));
	image.format = PNG_FORMAT_GRAY;
	int stride = PNG_IMAGE_ROW_STRIDE(image);
	int buf_size = PNG_IMAGE_SIZE(image);
	TEST_ASSERT_EQUAL(expected_width, image.width);
	TEST_ASSERT_EQUAL(expected_height, image.height);

[igrr]

Perhaps you can use a slightly more readable function

static inline uint16_t ldb_word(const void* ptr) {
  uint16_t result;
  memcpy(&result, ptr, sizeof(result);
  return result;
}

the compiler will optimize out memcpy replacing it with either an unaligned load or a pair of byte loads, depending on whether the CPU supports unaligned loads or not.

[tore-espressif]

@igrr I ended up with the following implementation to preserve the big-endian to little-endian transformation. All other comments have been incorporated!

static inline uint16_t ldb_word(const void *ptr)
{
    const uint8_t *p = (const uint8_t *)ptr;
    return ((uint16_t)p[0] << 8) | p[1];
}

[tore-espressif]

@igrr Thank you for the review, will update soon! EDIT: done