BlockSupplier API and block algorithm improvements

[tpietzsch]

(1)
Add new BlockSupplier default method

BlockSupplier<U> andThen(Function<BlockSupplier<T>, UnaryBlockOperator<T, U>> function)

that takes a Function that produces a UnaryBlockOperator which is then append to the current BlockSupplier. (The function can pick an operator based on the type and dimensions of the current BlockSupplier.)
This allows to write, e.g.,

import static net.imglib2.algorithm.blocks.convert.Convert.convert;
import static net.imglib2.algorithm.blocks.downsample.Downsample.downsample;
import static net.imglib2.view.fluent.RandomAccessibleIntervalView.Extension.mirrorDouble;

Img<UnsignedByteType> img;

BlockSupplier<FloatType> downsampled = BlockSupplier
		.of(img.view().extend(mirrorDouble()))
		.andThen(convert(new FloatType()))
		.andThen(downsample(Offset.HALF_PIXEL));

(2)
Refactoring and augment existing block algorithm (Convert, Downsample, Transform):

• Provide the above "operator factories".
• Add javadoc.
• Add downsampling method that averages over arbitrarily sized input blocks (for downsampling factor other than 2).
• Add no-op block operator that is used when converting to the same type (e.g., in generic code, you want any type T to be converted to FloatType. When at runtime T already is FloatType, that is a no-op now.)

(3)
Add TilingBlockSupplier and BlockSupplier.tile(int...) default method.
TilingBlockSupplier wraps a source BlockSupplier, and splits large copy requests into several smaller copy calls on the source BlockSupplier.

Example use cases:

• Computing large outputs (e. g. to write to N5 or wrap as ArrayImg) with operators that have better performance with smaller block sizes.
• Avoiding excessively large blocks when chaining multiple downsampling operators.

BlockSupplier<FloatType> downsampled = BlockSupplier
		.of(img.view().extend(mirrorDouble()))
		.andThen(convert(new FloatType()))
		.andThen(downsample(Offset.HALF_PIXEL))
		.tile(32);