Improvements to HEIC and AVIF decoders

[NicolasHug]

HEIC and AVIF decoders are now publicly available. Some improvements we can make are:

• Better torchscript support (meh...)
• support for decode_image()
• Robust file format detection logic (espcially for HEIC).

More of that is documented here:

vision/torchvision/io/image.py

Lines 386 to 402 in f7b1cfa

	# TODO_AVIF_HEIC: Better support for torchscript. Scripting decode_avif of
	# decode_heic currently fails, mainly because of the logic
	# _load_extra_decoders_once() (using global variables, try/except statements,
	# etc.).
	# The ops (torch.ops.extra_decoders_ns.decode_*) are otherwise torchscript-able,
	# and users who need torchscript can always just wrap those.
	# TODO_AVIF_HEIC: decode_image() should work for those. The key technical issue
	# we have here is that the format detection logic of decode_image() is
	# implemented in torchvision, and torchvision has zero knowledge of
	# torchvision-extra-decoders, so we cannot call the AVIF/HEIC C++ decoders
	# (those in torchvision-extra-decoders) from there.
	# A trivial check that could be done within torchvision would be to check the
	# file extension, if a path was passed. We could also just implement the
	# AVIF/HEIC detection logic in Python as a fallback, if the file detection
	# didn't find any format. In any case: properly determining whether a file is
	# HEIC is far from trivial, and relying on libmagic would probably be best

Torchscript support isnt' that important especially consider the op themselves support torchscript. Support in decode_image() is more important. Some update since the comment was written: there might be a way to call the C++ AVIF/HEIC decoders from within the C++ implementation of decode_image: simply calling the dispatcher like

  auto dispatcher = torch::Dispatcher::singleton();
  dispatcher.call("extra_decoders_ns::decode_heic", ...);

Need to investigate this.