Pass mimetype to serializers, and support mimetype parameters in registry

[danielballan]

Tiled "serializers" are functions which take a Python data structure as input and produce bytes as output, for transport over the network. The serializers currently look like this:

tiled/tiled/serialization/array.py

Lines 38 to 43 in 48bb130

	def serialize_csv(array, metadata):
	if array.ndim > 2:
	raise UnsupportedShape(array.shape)
	file = io.StringIO()
	numpy.savetxt(file, array, fmt="%s", delimiter=",")
	return file.getvalue().encode()

These get registered to a mimetype, potentially multiple of them.

tiled/tiled/serialization/array.py

Lines 46 to 48 in 48bb130

	serialization_registry.register("array", "text/csv", serialize_csv)
	serialization_registry.register("array", "text/x-comma-separated-values", serialize_csv)
	serialization_registry.register("array", "text/plain", serialize_csv)

I propose changing the signature to add mimetype as the first argument.

def f(mimetype, array, metadata):
    ...

This is a common pattern in callback registries. (We adopted the same pattern in caproto.) It enables the same callback to be subscribed to multiple channels and easily distinguish which one of them is calling. I see two use cases:

1. Serializers that can handle many types but need to do different things (like PIL). We currently handle this with a convenience function.
2. Support for mimetype parameters, such as text/csv;header=present and text/csv;header=absent. It would be convenient for serializer_csv to be called in both cases, and to be able to adjust its output correctly.

Also, in support of (2), I think this the registry should be tweaked to support optional mimetype parameters, such as text/csv;header=present. When the registry is passed a mimetype with parameters, it should perform dispatch based on the type without arguments (text/csv) but it should pass the full mimetype with arguments to the serializer, which can then parse the mimetype and do what it likes with the arguments.