Add the object length to object mappings

[teor2345]

The object retrieval code downloads one or two pieces to determine how long an object is. But the archiver can read the length of the object from the block data.

Including the object length in the block data would improve object fetching latency, because all the object's pieces could be downloaded at the same time.

As part of this change, mappings could use the offset of the first data byte, rather than the offset of the encoded length. This would save us 1 downloaded piece, in some rare edge cases.

See #3318 and #3362 for more background.

Implementation Strategy

The object mappings are initially generated by the runtime, then modified by the archiver. To apply to all blocks, the archiver should:

• add the length to the mapping
• update the offset and piece index to exclude the encoded object length bytes

Updating the runtime code would only add the length to mappings for new blocks.