Write full shards, revise DatasetCodec

[tpietzsch]

This PR adds new methods to DatasetAccess:

void writeShard(PositionValueAccess pva, DataBlock<T> dataBlock, int level) throws N5IOException;

DataBlock<T> readShard(PositionValueAccess pva, long[] shardGridPosition, int[] shardSizeInPixels, int level) throws N5IOException;

to read and write a shard as a "larger DataBlock"
This internally reads all the DataBlocks in the shard and re-assembles the data into a larger flattened array (and reverse of that for writing).
(This is in preparation for the "big renaming" when shard becomes DataBlock, and DataBlock becomes chunk.)

This PR also revises the buld readBlocks and writeBlocks methods, fixing TODOs that were still open, namely:

• readBlocks returns blocks in the same order as requested. The returned List<DataBlock> is in the same order as the requested List<long[]> of block grid positions. It will have null for blocks that don't exist. (This probably makes larger parts of Implement shardExists and readBlocksExist #187 #189 unnecessary, because readBlocksExists can just filter nulls from the resulting list.)
• writeBlocks checks whether all blocks in a (sub-)shard are written, and if so avoids reading and decoding the existing data.

Finally, this PR also revises generics on DatasetCodec and merges DatasetCodec into BlockCodec on creation, making special treatment in DefaultDatasetAccess unnecessary (i.e., DatasetAccess doesn't need to know about DatasetCodec at all).

[tpietzsch]

After our discussion, I changed/added the following:

• Revise DatasetCodec generics to DatasetCodec<S,T> with DataBlock<T> encode(DataBlock<S>) etc.
• DatasetAccess.readShard() always truncates at the dataset border.
• DatasetAccess.read/writeShard() for non-sharded datasets short-cut to just read/write a DataBlock without doing any copying.
• DatasetAccess.read/writeShard() handle duplicate block positions correctly.

I also rewrote the git history so that the individual commits make more sense, and put @cmhulbert shardExists commits back on top.

[cmhulbert]

I fixed the merge conflict with ShardTest but otherwise, happy to merge.