store tiny files inline in the index

[sourcefrog]

Files under about 100 bytes in length could be better stored directly in the index:

• the file content is potentially smaller than a hash referring to its block
• it avoids a separate IO for a data block when reading and writing, and probably still helps even after #66

There is a complication that the 0.6 index format is encoded in json, which can't directly include binary content. Options here seem to be:

• Only inline files whose content is a UTF-8 string. Many small files
• base-64 escape the content (not very nice)
• Wait to transition to a non-json index encoding, perhaps CBOR