Behavior when writing invalid `DataBlock`s

[bogovicj]

What should the behavior of an N5Writer be when attempting to write an invalid block.
Where a block is "invalid" if it is of the wrong size and/or type, see this test.

Behavior of n5-3.5.1:

• Writing a too-large block succeeds
• Writing a too-small block succeeds
• Writing a wrong-type block succeeds

Behavior of n5-4.0.0-alpha-1:

• Writing a too-large block succeeds
• Writing a too-small block succeeds
• Writing a wrong-type block succeeds

Behavior of the current wip/codecsReadData branch:

• Writing a too-large block succeeds
• Writing a too-small block succeeds
• Writing a wrong-type block fails with a ClassCastException

[tpietzsch]

Writing a wrong-type block fails with a ClassCastException

I think this makes more sense than the previous behavior

[bogovicj]

To clarify what I mean by too-small blocks:

1. Blocks whose data length are not consistent with the block size reported by the attributes.json metadata
2. Blocks whose data length are not consistent with the block size reported by the block header
3. Blocks with invalid locations

All of these write block calls write their invalid data without complaint:

DatasetAttributes attrs = DatasetAttributes.build(new long[] {5,5}, new int[] {5,5}, DataType.INT8);
n5.createDataset(dsetPath, attrs);

final ByteArrayDataBlock emptyBlockConflictingHeader = new ByteArrayDataBlock(new int[] {5, 5}, new long[]{0, 0}, new byte[] {});
n5.writeBlock(dsetPath, attrs, emptyBlockConflictingHeader);

final ByteArrayDataBlock emptyBlockConsistentHeader = new ByteArrayDataBlock(new int[] {0, 0}, new long[]{0, 0}, new byte[] {});
n5.writeBlock(dsetPath, attrs, emptyBlockConsistentHeader);

// this creates a an object at key "-1/-1/-1/-1/-1/-1"
final ByteArrayDataBlock blockWithCrazyLocation = new ByteArrayDataBlock(new int[] {0, 0}, new long[]{-1, -1, -1, -1, -1}, new byte[] {});
n5.writeBlock(dsetPath, attrs, blockWithCrazyLocation);

[bogovicj]

At this time, it may make sense to throw an exception when the DataBlock is created if an inconsistency is detected.

Checks we could consider adding:

• that a DataBlock's self reported size is consistent with the length of the provided data array
  • should perhaps be okay if the array is larger than needed
• that a DataBlock's location is non-negative for every dimension
• when calling writeBlock, that the DataBlock's size is consistent with the Dataset's block size
  • taking into account small blocks at the boundary for n5 datasets

[bogovicj]

@axtimwalde 's current preference is to allow the writing of blocks of the wrong type